program(1.3)
[buildInfo = dict<string, string>({{"coremlc-component-MIL", "3402.3.2"}, {"coremlc-version", "3402.4.1"}})]
{
    func infer<ios18>(tensor<fp16, [1, 1, 1, 2048]> causal_mask, tensor<int32, [1]> current_pos, tensor<fp16, [1, 1, 2048]> hidden_states, state<tensor<fp16, [32, 8, 2048, 64]>> model_model_kv_cache_0, tensor<int32, [1]> position_ids) {
            tensor<fp16, [2048, 2048, 1, 1]> model_model_layers_0_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64))), lut = tensor<fp16, [256, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2097280))))[name = string("model_model_layers_0_self_attn_q_proj_weight_palettized")];
            tensor<fp16, [512, 2048, 1, 1]> model_model_layers_0_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [512, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2105536))), lut = tensor<fp16, [64, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2629888))))[name = string("model_model_layers_0_self_attn_k_proj_weight_palettized")];
            tensor<fp16, [512, 2048, 1, 1]> model_model_layers_0_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [512, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2632000))), lut = tensor<fp16, [64, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3156352))))[name = string("model_model_layers_0_self_attn_v_proj_weight_palettized")];
            tensor<fp16, [8192, 2048, 1, 1]> model_model_layers_0_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [8192, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3158464))), lut = tensor<fp16, [1024, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11547136))))[name = string("model_model_layers_0_mlp_gate_proj_weight_palettized")];
            tensor<fp16, [8192, 2048, 1, 1]> model_model_layers_0_mlp_up_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [8192, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11579968))), lut = tensor<fp16, [1024, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19968640))))[name = string("model_model_layers_0_mlp_up_proj_weight_palettized")];
            tensor<fp16, [2048, 8192, 1, 1]> model_model_layers_0_mlp_down_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 8192, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20001472))), lut = tensor<fp16, [256, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28390144))))[name = string("model_model_layers_0_mlp_down_proj_weight_palettized")];
            tensor<fp16, [2048, 2048, 1, 1]> model_model_layers_1_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28398400))), lut = tensor<fp16, [256, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30495616))))[name = string("model_model_layers_1_self_attn_q_proj_weight_palettized")];
            tensor<fp16, [512, 2048, 1, 1]> model_model_layers_1_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [512, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30503872))), lut = tensor<fp16, [64, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31028224))))[name = string("model_model_layers_1_self_attn_k_proj_weight_palettized")];
            tensor<fp16, [512, 2048, 1, 1]> model_model_layers_1_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [512, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31030336))), lut = tensor<fp16, [64, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31554688))))[name = string("model_model_layers_1_self_attn_v_proj_weight_palettized")];
            tensor<fp16, [8192, 2048, 1, 1]> model_model_layers_1_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [8192, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31556800))), lut = tensor<fp16, [1024, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39945472))))[name = string("model_model_layers_1_mlp_gate_proj_weight_palettized")];
            tensor<fp16, [8192, 2048, 1, 1]> model_model_layers_1_mlp_up_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [8192, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39978304))), lut = tensor<fp16, [1024, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48366976))))[name = string("model_model_layers_1_mlp_up_proj_weight_palettized")];
            tensor<fp16, [2048, 8192, 1, 1]> model_model_layers_1_mlp_down_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 8192, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48399808))), lut = tensor<fp16, [256, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56788480))))[name = string("model_model_layers_1_mlp_down_proj_weight_palettized")];
            tensor<fp16, [2048, 2048, 1, 1]> model_model_layers_2_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56796736))), lut = tensor<fp16, [256, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58893952))))[name = string("model_model_layers_2_self_attn_q_proj_weight_palettized")];
            tensor<fp16, [512, 2048, 1, 1]> model_model_layers_2_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [512, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58902208))), lut = tensor<fp16, [64, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59426560))))[name = string("model_model_layers_2_self_attn_k_proj_weight_palettized")];
            tensor<fp16, [512, 2048, 1, 1]> model_model_layers_2_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [512, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59428672))), lut = tensor<fp16, [64, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59953024))))[name = string("model_model_layers_2_self_attn_v_proj_weight_palettized")];
            tensor<fp16, [8192, 2048, 1, 1]> model_model_layers_2_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [8192, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59955136))), lut = tensor<fp16, [1024, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68343808))))[name = string("model_model_layers_2_mlp_gate_proj_weight_palettized")];
            tensor<fp16, [8192, 2048, 1, 1]> model_model_layers_2_mlp_up_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [8192, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68376640))), lut = tensor<fp16, [1024, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76765312))))[name = string("model_model_layers_2_mlp_up_proj_weight_palettized")];
            tensor<fp16, [2048, 8192, 1, 1]> model_model_layers_2_mlp_down_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 8192, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76798144))), lut = tensor<fp16, [256, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85186816))))[name = string("model_model_layers_2_mlp_down_proj_weight_palettized")];
            tensor<fp16, [2048, 2048, 1, 1]> model_model_layers_3_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85195072))), lut = tensor<fp16, [256, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87292288))))[name = string("model_model_layers_3_self_attn_q_proj_weight_palettized")];
            tensor<fp16, [512, 2048, 1, 1]> model_model_layers_3_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [512, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87300544))), lut = tensor<fp16, [64, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87824896))))[name = string("model_model_layers_3_self_attn_k_proj_weight_palettized")];
            tensor<fp16, [512, 2048, 1, 1]> model_model_layers_3_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [512, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87827008))), lut = tensor<fp16, [64, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88351360))))[name = string("model_model_layers_3_self_attn_v_proj_weight_palettized")];
            tensor<fp16, [8192, 2048, 1, 1]> model_model_layers_3_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [8192, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88353472))), lut = tensor<fp16, [1024, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96742144))))[name = string("model_model_layers_3_mlp_gate_proj_weight_palettized")];
            tensor<fp16, [8192, 2048, 1, 1]> model_model_layers_3_mlp_up_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [8192, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96774976))), lut = tensor<fp16, [1024, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105163648))))[name = string("model_model_layers_3_mlp_up_proj_weight_palettized")];
            tensor<fp16, [2048, 8192, 1, 1]> model_model_layers_3_mlp_down_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 8192, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105196480))), lut = tensor<fp16, [256, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113585152))))[name = string("model_model_layers_3_mlp_down_proj_weight_palettized")];
            tensor<fp16, [2048, 2048, 1, 1]> model_model_layers_4_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113593408))), lut = tensor<fp16, [256, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115690624))))[name = string("model_model_layers_4_self_attn_q_proj_weight_palettized")];
            tensor<fp16, [512, 2048, 1, 1]> model_model_layers_4_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [512, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115698880))), lut = tensor<fp16, [64, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116223232))))[name = string("model_model_layers_4_self_attn_k_proj_weight_palettized")];
            tensor<fp16, [512, 2048, 1, 1]> model_model_layers_4_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [512, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116225344))), lut = tensor<fp16, [64, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116749696))))[name = string("model_model_layers_4_self_attn_v_proj_weight_palettized")];
            tensor<fp16, [8192, 2048, 1, 1]> model_model_layers_4_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [8192, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116751808))), lut = tensor<fp16, [1024, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125140480))))[name = string("model_model_layers_4_mlp_gate_proj_weight_palettized")];
            tensor<fp16, [8192, 2048, 1, 1]> model_model_layers_4_mlp_up_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [8192, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125173312))), lut = tensor<fp16, [1024, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133561984))))[name = string("model_model_layers_4_mlp_up_proj_weight_palettized")];
            tensor<fp16, [2048, 8192, 1, 1]> model_model_layers_4_mlp_down_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 8192, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133594816))), lut = tensor<fp16, [256, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141983488))))[name = string("model_model_layers_4_mlp_down_proj_weight_palettized")];
            tensor<fp16, [2048, 2048, 1, 1]> model_model_layers_5_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141991744))), lut = tensor<fp16, [256, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144088960))))[name = string("model_model_layers_5_self_attn_q_proj_weight_palettized")];
            tensor<fp16, [512, 2048, 1, 1]> model_model_layers_5_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [512, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144097216))), lut = tensor<fp16, [64, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144621568))))[name = string("model_model_layers_5_self_attn_k_proj_weight_palettized")];
            tensor<fp16, [512, 2048, 1, 1]> model_model_layers_5_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [512, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144623680))), lut = tensor<fp16, [64, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145148032))))[name = string("model_model_layers_5_self_attn_v_proj_weight_palettized")];
            tensor<fp16, [8192, 2048, 1, 1]> model_model_layers_5_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [8192, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145150144))), lut = tensor<fp16, [1024, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153538816))))[name = string("model_model_layers_5_mlp_gate_proj_weight_palettized")];
            tensor<fp16, [8192, 2048, 1, 1]> model_model_layers_5_mlp_up_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [8192, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153571648))), lut = tensor<fp16, [1024, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161960320))))[name = string("model_model_layers_5_mlp_up_proj_weight_palettized")];
            tensor<fp16, [2048, 8192, 1, 1]> model_model_layers_5_mlp_down_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 8192, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161993152))), lut = tensor<fp16, [256, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170381824))))[name = string("model_model_layers_5_mlp_down_proj_weight_palettized")];
            tensor<fp16, [2048, 2048, 1, 1]> model_model_layers_6_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170390080))), lut = tensor<fp16, [256, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172487296))))[name = string("model_model_layers_6_self_attn_q_proj_weight_palettized")];
            tensor<fp16, [512, 2048, 1, 1]> model_model_layers_6_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [512, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172495552))), lut = tensor<fp16, [64, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173019904))))[name = string("model_model_layers_6_self_attn_k_proj_weight_palettized")];
            tensor<fp16, [512, 2048, 1, 1]> model_model_layers_6_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [512, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173022016))), lut = tensor<fp16, [64, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173546368))))[name = string("model_model_layers_6_self_attn_v_proj_weight_palettized")];
            tensor<fp16, [8192, 2048, 1, 1]> model_model_layers_6_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [8192, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173548480))), lut = tensor<fp16, [1024, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181937152))))[name = string("model_model_layers_6_mlp_gate_proj_weight_palettized")];
            tensor<fp16, [8192, 2048, 1, 1]> model_model_layers_6_mlp_up_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [8192, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181969984))), lut = tensor<fp16, [1024, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190358656))))[name = string("model_model_layers_6_mlp_up_proj_weight_palettized")];
            tensor<fp16, [2048, 8192, 1, 1]> model_model_layers_6_mlp_down_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 8192, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190391488))), lut = tensor<fp16, [256, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198780160))))[name = string("model_model_layers_6_mlp_down_proj_weight_palettized")];
            tensor<fp16, [2048, 2048, 1, 1]> model_model_layers_7_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198788416))), lut = tensor<fp16, [256, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200885632))))[name = string("model_model_layers_7_self_attn_q_proj_weight_palettized")];
            tensor<fp16, [512, 2048, 1, 1]> model_model_layers_7_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [512, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200893888))), lut = tensor<fp16, [64, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201418240))))[name = string("model_model_layers_7_self_attn_k_proj_weight_palettized")];
            tensor<fp16, [512, 2048, 1, 1]> model_model_layers_7_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [512, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201420352))), lut = tensor<fp16, [64, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201944704))))[name = string("model_model_layers_7_self_attn_v_proj_weight_palettized")];
            tensor<fp16, [8192, 2048, 1, 1]> model_model_layers_7_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [8192, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201946816))), lut = tensor<fp16, [1024, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210335488))))[name = string("model_model_layers_7_mlp_gate_proj_weight_palettized")];
            tensor<fp16, [8192, 2048, 1, 1]> model_model_layers_7_mlp_up_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [8192, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210368320))), lut = tensor<fp16, [1024, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(218756992))))[name = string("model_model_layers_7_mlp_up_proj_weight_palettized")];
            tensor<fp16, [2048, 8192, 1, 1]> model_model_layers_7_mlp_down_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 8192, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(218789824))), lut = tensor<fp16, [256, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227178496))))[name = string("model_model_layers_7_mlp_down_proj_weight_palettized")];
            int32 var_49 = const()[name = string("op_49"), val = int32(-1)];
            int32 greater_equal_0_y_0 = const()[name = string("greater_equal_0_y_0"), val = int32(0)];
            tensor<bool, [1]> greater_equal_0 = greater_equal(x = current_pos, y = greater_equal_0_y_0)[name = string("greater_equal_0")];
            int32 slice_by_index_0 = const()[name = string("slice_by_index_0"), val = int32(131072)];
            tensor<int32, [1]> add_0 = add(x = current_pos, y = slice_by_index_0)[name = string("add_0")];
            tensor<int32, [1]> select_0 = select(a = current_pos, b = add_0, cond = greater_equal_0)[name = string("select_0")];
            int32 var_258_axis_0 = const()[name = string("op_258_axis_0"), val = int32(1)];
            int32 var_258_batch_dims_0 = const()[name = string("op_258_batch_dims_0"), val = int32(0)];
            bool var_258_validate_indices_0 = const()[name = string("op_258_validate_indices_0"), val = bool(false)];
            tensor<fp16, [1, 131072, 64]> var_54_to_fp16 = const()[name = string("op_54_to_fp16"), val = tensor<fp16, [1, 131072, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227186752)))];
            tensor<fp16, [1, 1, 64]> var_258_cast_fp16 = gather(axis = var_258_axis_0, batch_dims = var_258_batch_dims_0, indices = select_0, validate_indices = var_258_validate_indices_0, x = var_54_to_fp16)[name = string("op_258_cast_fp16")];
            tensor<int32, [4]> var_259 = const()[name = string("op_259"), val = tensor<int32, [4]>([1, 1, 1, -1])];
            tensor<fp16, [1, 1, 1, 64]> sin_1_cast_fp16 = reshape(shape = var_259, x = var_258_cast_fp16)[name = string("sin_1_cast_fp16")];
            int32 var_263_axis_0 = const()[name = string("op_263_axis_0"), val = int32(1)];
            int32 var_263_batch_dims_0 = const()[name = string("op_263_batch_dims_0"), val = int32(0)];
            bool var_263_validate_indices_0 = const()[name = string("op_263_validate_indices_0"), val = bool(false)];
            tensor<fp16, [1, 131072, 64]> var_48_to_fp16 = const()[name = string("op_48_to_fp16"), val = tensor<fp16, [1, 131072, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243964032)))];
            tensor<fp16, [1, 1, 64]> var_263_cast_fp16 = gather(axis = var_263_axis_0, batch_dims = var_263_batch_dims_0, indices = select_0, validate_indices = var_263_validate_indices_0, x = var_48_to_fp16)[name = string("op_263_cast_fp16")];
            tensor<int32, [4]> var_264 = const()[name = string("op_264"), val = tensor<int32, [4]>([1, 1, 1, -1])];
            tensor<fp16, [1, 1, 1, 64]> cos_1_cast_fp16 = reshape(shape = var_264, x = var_263_cast_fp16)[name = string("cos_1_cast_fp16")];
            tensor<int32, [1]> mean_1_axes_0 = const()[name = string("mean_1_axes_0"), val = tensor<int32, [1]>([-1])];
            bool mean_1_keep_dims_0 = const()[name = string("mean_1_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 1, 1]> mean_1_cast_fp16 = reduce_mean(axes = mean_1_axes_0, keep_dims = mean_1_keep_dims_0, x = hidden_states)[name = string("mean_1_cast_fp16")];
            tensor<fp16, [1, 1, 2048]> input_1_cast_fp16 = sub(x = hidden_states, y = mean_1_cast_fp16)[name = string("input_1_cast_fp16")];
            tensor<int32, [1]> var_272_axes_0 = const()[name = string("op_272_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2048]> model_model_layers_0_input_layernorm_weight_to_fp16 = const()[name = string("model_model_layers_0_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260741312)))];
            fp16 var_44_to_fp16 = const()[name = string("op_44_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1, 2048]> var_272_cast_fp16 = layer_norm(axes = var_272_axes_0, epsilon = var_44_to_fp16, gamma = model_model_layers_0_input_layernorm_weight_to_fp16, x = input_1_cast_fp16)[name = string("op_272_cast_fp16")];
            tensor<int32, [3]> var_275 = const()[name = string("op_275"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_277_axes_0 = const()[name = string("op_277_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 1]> var_276 = transpose(perm = var_275, x = var_272_cast_fp16)[name = string("transpose_31")];
            tensor<fp16, [1, 2048, 1, 1]> var_277 = expand_dims(axes = var_277_axes_0, x = var_276)[name = string("op_277")];
            string var_284_pad_type_0 = const()[name = string("op_284_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_284_strides_0 = const()[name = string("op_284_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_284_pad_0 = const()[name = string("op_284_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_284_dilations_0 = const()[name = string("op_284_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_284_groups_0 = const()[name = string("op_284_groups_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 1, 1]> var_284 = conv(dilations = var_284_dilations_0, groups = var_284_groups_0, pad = var_284_pad_0, pad_type = var_284_pad_type_0, strides = var_284_strides_0, weight = model_model_layers_0_self_attn_q_proj_weight_palettized, x = var_277)[name = string("op_284")];
            tensor<int32, [4]> var_285 = const()[name = string("op_285"), val = tensor<int32, [4]>([1, 32, 1, 64])];
            tensor<fp16, [1, 32, 1, 64]> var_286 = reshape(shape = var_285, x = var_284)[name = string("op_286")];
            string var_293_pad_type_0 = const()[name = string("op_293_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_293_strides_0 = const()[name = string("op_293_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_293_pad_0 = const()[name = string("op_293_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_293_dilations_0 = const()[name = string("op_293_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_293_groups_0 = const()[name = string("op_293_groups_0"), val = int32(1)];
            tensor<fp16, [1, 512, 1, 1]> var_293 = conv(dilations = var_293_dilations_0, groups = var_293_groups_0, pad = var_293_pad_0, pad_type = var_293_pad_type_0, strides = var_293_strides_0, weight = model_model_layers_0_self_attn_k_proj_weight_palettized, x = var_277)[name = string("op_293")];
            tensor<int32, [4]> var_294 = const()[name = string("op_294"), val = tensor<int32, [4]>([1, 8, 1, 64])];
            tensor<fp16, [1, 8, 1, 64]> var_295 = reshape(shape = var_294, x = var_293)[name = string("op_295")];
            string var_302_pad_type_0 = const()[name = string("op_302_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_302_strides_0 = const()[name = string("op_302_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_302_pad_0 = const()[name = string("op_302_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_302_dilations_0 = const()[name = string("op_302_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_302_groups_0 = const()[name = string("op_302_groups_0"), val = int32(1)];
            tensor<fp16, [1, 512, 1, 1]> var_302 = conv(dilations = var_302_dilations_0, groups = var_302_groups_0, pad = var_302_pad_0, pad_type = var_302_pad_type_0, strides = var_302_strides_0, weight = model_model_layers_0_self_attn_v_proj_weight_palettized, x = var_277)[name = string("op_302")];
            tensor<int32, [4]> var_303 = const()[name = string("op_303"), val = tensor<int32, [4]>([1, 8, 1, 64])];
            tensor<fp16, [1, 8, 1, 64]> var_304 = reshape(shape = var_303, x = var_302)[name = string("op_304")];
            tensor<int32, [4]> x1_1_begin_0 = const()[name = string("x1_1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_1_end_0 = const()[name = string("x1_1_end_0"), val = tensor<int32, [4]>([1, 32, 1, 32])];
            tensor<bool, [4]> x1_1_end_mask_0 = const()[name = string("x1_1_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 1, 32]> x1_1 = slice_by_index(begin = x1_1_begin_0, end = x1_1_end_0, end_mask = x1_1_end_mask_0, x = var_286)[name = string("x1_1")];
            tensor<int32, [4]> x2_1_begin_0 = const()[name = string("x2_1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_1_end_0 = const()[name = string("x2_1_end_0"), val = tensor<int32, [4]>([1, 32, 1, 64])];
            tensor<bool, [4]> x2_1_end_mask_0 = const()[name = string("x2_1_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 1, 32]> x2_1 = slice_by_index(begin = x2_1_begin_0, end = x2_1_end_0, end_mask = x2_1_end_mask_0, x = var_286)[name = string("x2_1")];
            tensor<int32, [4]> cos_3_begin_0 = const()[name = string("cos_3_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> cos_3_end_0 = const()[name = string("cos_3_end_0"), val = tensor<int32, [4]>([1, 1, 1, 32])];
            tensor<bool, [4]> cos_3_end_mask_0 = const()[name = string("cos_3_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1, 1, 32]> cos_3_cast_fp16 = slice_by_index(begin = cos_3_begin_0, end = cos_3_end_0, end_mask = cos_3_end_mask_0, x = cos_1_cast_fp16)[name = string("cos_3_cast_fp16")];
            tensor<int32, [4]> sin_3_begin_0 = const()[name = string("sin_3_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> sin_3_end_0 = const()[name = string("sin_3_end_0"), val = tensor<int32, [4]>([1, 1, 1, 32])];
            tensor<bool, [4]> sin_3_end_mask_0 = const()[name = string("sin_3_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1, 1, 32]> sin_3_cast_fp16 = slice_by_index(begin = sin_3_begin_0, end = sin_3_end_0, end_mask = sin_3_end_mask_0, x = sin_1_cast_fp16)[name = string("sin_3_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_318_cast_fp16 = mul(x = x1_1, y = cos_3_cast_fp16)[name = string("op_318_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_319_cast_fp16 = mul(x = x2_1, y = sin_3_cast_fp16)[name = string("op_319_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_320_cast_fp16 = sub(x = var_318_cast_fp16, y = var_319_cast_fp16)[name = string("op_320_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_321_cast_fp16 = mul(x = x2_1, y = cos_3_cast_fp16)[name = string("op_321_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_322_cast_fp16 = mul(x = x1_1, y = sin_3_cast_fp16)[name = string("op_322_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_323_cast_fp16 = add(x = var_321_cast_fp16, y = var_322_cast_fp16)[name = string("op_323_cast_fp16")];
            bool rotated_1_interleave_0 = const()[name = string("rotated_1_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 32, 1, 64]> rotated_1_cast_fp16 = concat(axis = var_49, interleave = rotated_1_interleave_0, values = (var_320_cast_fp16, var_323_cast_fp16))[name = string("rotated_1_cast_fp16")];
            tensor<int32, [4]> x1_3_begin_0 = const()[name = string("x1_3_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_3_end_0 = const()[name = string("x1_3_end_0"), val = tensor<int32, [4]>([1, 8, 1, 32])];
            tensor<bool, [4]> x1_3_end_mask_0 = const()[name = string("x1_3_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 1, 32]> x1_3 = slice_by_index(begin = x1_3_begin_0, end = x1_3_end_0, end_mask = x1_3_end_mask_0, x = var_295)[name = string("x1_3")];
            tensor<int32, [4]> x2_3_begin_0 = const()[name = string("x2_3_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_3_end_0 = const()[name = string("x2_3_end_0"), val = tensor<int32, [4]>([1, 8, 1, 64])];
            tensor<bool, [4]> x2_3_end_mask_0 = const()[name = string("x2_3_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 1, 32]> x2_3 = slice_by_index(begin = x2_3_begin_0, end = x2_3_end_0, end_mask = x2_3_end_mask_0, x = var_295)[name = string("x2_3")];
            tensor<fp16, [1, 8, 1, 32]> var_339_cast_fp16 = mul(x = x1_3, y = cos_3_cast_fp16)[name = string("op_339_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_340_cast_fp16 = mul(x = x2_3, y = sin_3_cast_fp16)[name = string("op_340_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_341_cast_fp16 = sub(x = var_339_cast_fp16, y = var_340_cast_fp16)[name = string("op_341_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_342_cast_fp16 = mul(x = x2_3, y = cos_3_cast_fp16)[name = string("op_342_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_343_cast_fp16 = mul(x = x1_3, y = sin_3_cast_fp16)[name = string("op_343_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_344_cast_fp16 = add(x = var_342_cast_fp16, y = var_343_cast_fp16)[name = string("op_344_cast_fp16")];
            bool rotated_3_interleave_0 = const()[name = string("rotated_3_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 8, 1, 64]> rotated_3_cast_fp16 = concat(axis = var_49, interleave = rotated_3_interleave_0, values = (var_341_cast_fp16, var_344_cast_fp16))[name = string("rotated_3_cast_fp16")];
            int32 var_348 = const()[name = string("op_348"), val = int32(1)];
            tensor<int32, [1]> var_349 = add(x = current_pos, y = var_348)[name = string("op_349")];
            tensor<fp16, [32, 8, 2048, 64]> read_state_0 = read_state(input = model_model_kv_cache_0)[name = string("read_state_0")];
            tensor<int32, [1]> expand_dims_0 = const()[name = string("expand_dims_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_1 = const()[name = string("expand_dims_1"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_3 = const()[name = string("expand_dims_3"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_4 = const()[name = string("expand_dims_4"), val = tensor<int32, [1]>([1])];
            int32 concat_2_axis_0 = const()[name = string("concat_2_axis_0"), val = int32(0)];
            bool concat_2_interleave_0 = const()[name = string("concat_2_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_2 = concat(axis = concat_2_axis_0, interleave = concat_2_interleave_0, values = (expand_dims_0, expand_dims_1, current_pos, expand_dims_3))[name = string("concat_2")];
            tensor<int32, [1]> concat_3_values1_0 = const()[name = string("concat_3_values1_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> concat_3_values3_0 = const()[name = string("concat_3_values3_0"), val = tensor<int32, [1]>([0])];
            int32 concat_3_axis_0 = const()[name = string("concat_3_axis_0"), val = int32(0)];
            bool concat_3_interleave_0 = const()[name = string("concat_3_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_3 = concat(axis = concat_3_axis_0, interleave = concat_3_interleave_0, values = (expand_dims_4, concat_3_values1_0, var_349, concat_3_values3_0))[name = string("concat_3")];
            tensor<int32, [4]> model_model_kv_cache_0_internal_tensor_assign_1_stride_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_1_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_1_begin_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_1_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_1_end_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_1_end_mask_0"), val = tensor<bool, [4]>([false, true, false, true])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_1_squeeze_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_1_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<fp16, [32, 8, 2048, 64]> model_model_kv_cache_0_internal_tensor_assign_1_cast_fp16 = slice_update(begin = concat_2, begin_mask = model_model_kv_cache_0_internal_tensor_assign_1_begin_mask_0, end = concat_3, end_mask = model_model_kv_cache_0_internal_tensor_assign_1_end_mask_0, squeeze_mask = model_model_kv_cache_0_internal_tensor_assign_1_squeeze_mask_0, stride = model_model_kv_cache_0_internal_tensor_assign_1_stride_0, update = rotated_3_cast_fp16, x = read_state_0)[name = string("model_model_kv_cache_0_internal_tensor_assign_1_cast_fp16")];
            write_state(data = model_model_kv_cache_0_internal_tensor_assign_1_cast_fp16, input = model_model_kv_cache_0)[name = string("coreml_update_state_0_write_state")];
            tensor<fp16, [32, 8, 2048, 64]> coreml_update_state_16 = read_state(input = model_model_kv_cache_0)[name = string("coreml_update_state_0")];
            tensor<int32, [1]> expand_dims_6 = const()[name = string("expand_dims_6"), val = tensor<int32, [1]>([16])];
            tensor<int32, [1]> expand_dims_7 = const()[name = string("expand_dims_7"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_9 = const()[name = string("expand_dims_9"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_10 = const()[name = string("expand_dims_10"), val = tensor<int32, [1]>([17])];
            int32 concat_6_axis_0 = const()[name = string("concat_6_axis_0"), val = int32(0)];
            bool concat_6_interleave_0 = const()[name = string("concat_6_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_6 = concat(axis = concat_6_axis_0, interleave = concat_6_interleave_0, values = (expand_dims_6, expand_dims_7, current_pos, expand_dims_9))[name = string("concat_6")];
            tensor<int32, [1]> concat_7_values1_0 = const()[name = string("concat_7_values1_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> concat_7_values3_0 = const()[name = string("concat_7_values3_0"), val = tensor<int32, [1]>([0])];
            int32 concat_7_axis_0 = const()[name = string("concat_7_axis_0"), val = int32(0)];
            bool concat_7_interleave_0 = const()[name = string("concat_7_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_7 = concat(axis = concat_7_axis_0, interleave = concat_7_interleave_0, values = (expand_dims_10, concat_7_values1_0, var_349, concat_7_values3_0))[name = string("concat_7")];
            tensor<int32, [4]> model_model_kv_cache_0_internal_tensor_assign_2_stride_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_2_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_2_begin_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_2_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_2_end_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_2_end_mask_0"), val = tensor<bool, [4]>([false, true, false, true])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_2_squeeze_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_2_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<fp16, [32, 8, 2048, 64]> model_model_kv_cache_0_internal_tensor_assign_2_cast_fp16 = slice_update(begin = concat_6, begin_mask = model_model_kv_cache_0_internal_tensor_assign_2_begin_mask_0, end = concat_7, end_mask = model_model_kv_cache_0_internal_tensor_assign_2_end_mask_0, squeeze_mask = model_model_kv_cache_0_internal_tensor_assign_2_squeeze_mask_0, stride = model_model_kv_cache_0_internal_tensor_assign_2_stride_0, update = var_304, x = coreml_update_state_16)[name = string("model_model_kv_cache_0_internal_tensor_assign_2_cast_fp16")];
            write_state(data = model_model_kv_cache_0_internal_tensor_assign_2_cast_fp16, input = model_model_kv_cache_0)[name = string("coreml_update_state_1_write_state")];
            tensor<fp16, [32, 8, 2048, 64]> coreml_update_state_17 = read_state(input = model_model_kv_cache_0)[name = string("coreml_update_state_1")];
            tensor<int32, [4]> var_364_begin_0 = const()[name = string("op_364_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_364_end_0 = const()[name = string("op_364_end_0"), val = tensor<int32, [4]>([1, 8, 2048, 64])];
            tensor<bool, [4]> var_364_end_mask_0 = const()[name = string("op_364_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<fp16, [1, 8, 2048, 64]> var_364_cast_fp16 = slice_by_index(begin = var_364_begin_0, end = var_364_end_0, end_mask = var_364_end_mask_0, x = coreml_update_state_17)[name = string("op_364_cast_fp16")];
            tensor<int32, [1]> K_layer_cache_1_axes_0 = const()[name = string("K_layer_cache_1_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<fp16, [8, 2048, 64]> K_layer_cache_1_cast_fp16 = squeeze(axes = K_layer_cache_1_axes_0, x = var_364_cast_fp16)[name = string("K_layer_cache_1_cast_fp16")];
            tensor<int32, [4]> var_366_begin_0 = const()[name = string("op_366_begin_0"), val = tensor<int32, [4]>([16, 0, 0, 0])];
            tensor<int32, [4]> var_366_end_0 = const()[name = string("op_366_end_0"), val = tensor<int32, [4]>([17, 8, 2048, 64])];
            tensor<bool, [4]> var_366_end_mask_0 = const()[name = string("op_366_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<fp16, [1, 8, 2048, 64]> var_366_cast_fp16 = slice_by_index(begin = var_366_begin_0, end = var_366_end_0, end_mask = var_366_end_mask_0, x = coreml_update_state_17)[name = string("op_366_cast_fp16")];
            tensor<int32, [1]> V_layer_cache_1_axes_0 = const()[name = string("V_layer_cache_1_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<fp16, [8, 2048, 64]> V_layer_cache_1_cast_fp16 = squeeze(axes = V_layer_cache_1_axes_0, x = var_366_cast_fp16)[name = string("V_layer_cache_1_cast_fp16")];
            tensor<int32, [1]> x_11_axes_0 = const()[name = string("x_11_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [8, 1, 2048, 64]> x_11_cast_fp16 = expand_dims(axes = x_11_axes_0, x = K_layer_cache_1_cast_fp16)[name = string("x_11_cast_fp16")];
            tensor<int32, [4]> var_375 = const()[name = string("op_375"), val = tensor<int32, [4]>([1, 4, 1, 1])];
            tensor<fp16, [8, 4, 2048, 64]> x_13_cast_fp16 = tile(reps = var_375, x = x_11_cast_fp16)[name = string("x_13_cast_fp16")];
            tensor<int32, [4]> var_379 = const()[name = string("op_379"), val = tensor<int32, [4]>([1, -1, 2048, 64])];
            tensor<fp16, [1, 32, 2048, 64]> key_states_3_cast_fp16 = reshape(shape = var_379, x = x_13_cast_fp16)[name = string("key_states_3_cast_fp16")];
            tensor<int32, [1]> x_17_axes_0 = const()[name = string("x_17_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [8, 1, 2048, 64]> x_17_cast_fp16 = expand_dims(axes = x_17_axes_0, x = V_layer_cache_1_cast_fp16)[name = string("x_17_cast_fp16")];
            tensor<int32, [4]> var_382 = const()[name = string("op_382"), val = tensor<int32, [4]>([1, 4, 1, 1])];
            tensor<fp16, [8, 4, 2048, 64]> x_19_cast_fp16 = tile(reps = var_382, x = x_17_cast_fp16)[name = string("x_19_cast_fp16")];
            tensor<int32, [4]> var_386 = const()[name = string("op_386"), val = tensor<int32, [4]>([1, -1, 2048, 64])];
            tensor<fp16, [1, 32, 2048, 64]> value_states_3_cast_fp16 = reshape(shape = var_386, x = x_19_cast_fp16)[name = string("value_states_3_cast_fp16")];
            bool var_389_transpose_x_1 = const()[name = string("op_389_transpose_x_1"), val = bool(false)];
            bool var_389_transpose_y_1 = const()[name = string("op_389_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 2048]> var_389_cast_fp16 = matmul(transpose_x = var_389_transpose_x_1, transpose_y = var_389_transpose_y_1, x = rotated_1_cast_fp16, y = key_states_3_cast_fp16)[name = string("op_389_cast_fp16")];
            fp16 var_390_to_fp16 = const()[name = string("op_390_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 32, 1, 2048]> attn_weights_1_cast_fp16 = mul(x = var_389_cast_fp16, y = var_390_to_fp16)[name = string("attn_weights_1_cast_fp16")];
            tensor<fp16, [1, 32, 1, 2048]> x_21_cast_fp16 = add(x = attn_weights_1_cast_fp16, y = causal_mask)[name = string("x_21_cast_fp16")];
            tensor<int32, [1]> reduce_max_0_axes_0 = const()[name = string("reduce_max_0_axes_0"), val = tensor<int32, [1]>([-1])];
            bool reduce_max_0_keep_dims_0 = const()[name = string("reduce_max_0_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1]> reduce_max_0_cast_fp16 = reduce_max(axes = reduce_max_0_axes_0, keep_dims = reduce_max_0_keep_dims_0, x = x_21_cast_fp16)[name = string("reduce_max_0_cast_fp16")];
            tensor<fp16, [1, 32, 1, 2048]> x_23_cast_fp16 = sub(x = x_21_cast_fp16, y = reduce_max_0_cast_fp16)[name = string("x_23_cast_fp16")];
            tensor<fp16, [1, 32, 1, 2048]> exp_x_1_cast_fp16 = exp(x = x_23_cast_fp16)[name = string("exp_x_1_cast_fp16")];
            tensor<int32, [1]> var_401_axes_0 = const()[name = string("op_401_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_401_keep_dims_0 = const()[name = string("op_401_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1]> var_401_cast_fp16 = reduce_sum(axes = var_401_axes_0, keep_dims = var_401_keep_dims_0, x = exp_x_1_cast_fp16)[name = string("op_401_cast_fp16")];
            tensor<fp16, [1, 32, 1, 2048]> attn_weights_3_cast_fp16 = real_div(x = exp_x_1_cast_fp16, y = var_401_cast_fp16)[name = string("attn_weights_3_cast_fp16")];
            bool attn_output_1_transpose_x_0 = const()[name = string("attn_output_1_transpose_x_0"), val = bool(false)];
            bool attn_output_1_transpose_y_0 = const()[name = string("attn_output_1_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 32, 1, 64]> attn_output_1_cast_fp16 = matmul(transpose_x = attn_output_1_transpose_x_0, transpose_y = attn_output_1_transpose_y_0, x = attn_weights_3_cast_fp16, y = value_states_3_cast_fp16)[name = string("attn_output_1_cast_fp16")];
            tensor<int32, [4]> var_404_perm_0 = const()[name = string("op_404_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_406 = const()[name = string("op_406"), val = tensor<int32, [3]>([1, 1, 2048])];
            tensor<fp16, [1, 1, 32, 64]> var_404_cast_fp16 = transpose(perm = var_404_perm_0, x = attn_output_1_cast_fp16)[name = string("transpose_30")];
            tensor<fp16, [1, 1, 2048]> input_5_cast_fp16 = reshape(shape = var_406, x = var_404_cast_fp16)[name = string("input_5_cast_fp16")];
            tensor<fp16, [2048, 2048]> model_model_layers_0_self_attn_o_proj_weight_promoted_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260745472))), lut = tensor<fp16, [256, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262842688))))[name = string("model_model_layers_0_self_attn_o_proj_weight_promoted_to_fp16_palettized")];
            tensor<fp16, [2048]> linear_0_bias_0_to_fp16 = const()[name = string("linear_0_bias_0_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262850944)))];
            tensor<fp16, [1, 1, 2048]> linear_0_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_0_self_attn_o_proj_weight_promoted_to_fp16_palettized, x = input_5_cast_fp16)[name = string("linear_0_cast_fp16")];
            tensor<fp16, [1, 1, 2048]> hidden_states_5_cast_fp16 = add(x = hidden_states, y = linear_0_cast_fp16)[name = string("hidden_states_5_cast_fp16")];
            tensor<int32, [1]> mean_3_axes_0 = const()[name = string("mean_3_axes_0"), val = tensor<int32, [1]>([-1])];
            bool mean_3_keep_dims_0 = const()[name = string("mean_3_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 1, 1]> mean_3_cast_fp16 = reduce_mean(axes = mean_3_axes_0, keep_dims = mean_3_keep_dims_0, x = hidden_states_5_cast_fp16)[name = string("mean_3_cast_fp16")];
            tensor<fp16, [1, 1, 2048]> input_7_cast_fp16 = sub(x = hidden_states_5_cast_fp16, y = mean_3_cast_fp16)[name = string("input_7_cast_fp16")];
            tensor<int32, [1]> var_417_axes_0 = const()[name = string("op_417_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2048]> model_model_layers_0_post_attention_layernorm_weight_to_fp16 = const()[name = string("model_model_layers_0_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262855104)))];
            tensor<fp16, [1, 1, 2048]> var_417_cast_fp16 = layer_norm(axes = var_417_axes_0, epsilon = var_44_to_fp16, gamma = model_model_layers_0_post_attention_layernorm_weight_to_fp16, x = input_7_cast_fp16)[name = string("op_417_cast_fp16")];
            tensor<int32, [3]> var_424 = const()[name = string("op_424"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> input_9_axes_0 = const()[name = string("input_9_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 1]> var_425 = transpose(perm = var_424, x = var_417_cast_fp16)[name = string("transpose_29")];
            tensor<fp16, [1, 2048, 1, 1]> input_9 = expand_dims(axes = input_9_axes_0, x = var_425)[name = string("input_9")];
            string input_11_pad_type_0 = const()[name = string("input_11_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_11_strides_0 = const()[name = string("input_11_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_11_pad_0 = const()[name = string("input_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_11_dilations_0 = const()[name = string("input_11_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_11_groups_0 = const()[name = string("input_11_groups_0"), val = int32(1)];
            tensor<fp16, [1, 8192, 1, 1]> input_11 = conv(dilations = input_11_dilations_0, groups = input_11_groups_0, pad = input_11_pad_0, pad_type = input_11_pad_type_0, strides = input_11_strides_0, weight = model_model_layers_0_mlp_gate_proj_weight_palettized, x = input_9)[name = string("input_11")];
            string up_states_1_pad_type_0 = const()[name = string("up_states_1_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> up_states_1_strides_0 = const()[name = string("up_states_1_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> up_states_1_pad_0 = const()[name = string("up_states_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> up_states_1_dilations_0 = const()[name = string("up_states_1_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 up_states_1_groups_0 = const()[name = string("up_states_1_groups_0"), val = int32(1)];
            tensor<fp16, [1, 8192, 1, 1]> up_states_1 = conv(dilations = up_states_1_dilations_0, groups = up_states_1_groups_0, pad = up_states_1_pad_0, pad_type = up_states_1_pad_type_0, strides = up_states_1_strides_0, weight = model_model_layers_0_mlp_up_proj_weight_palettized, x = input_9)[name = string("up_states_1")];
            tensor<fp16, [1, 8192, 1, 1]> gate_states_1 = silu(x = input_11)[name = string("gate_states_1")];
            tensor<fp16, [1, 8192, 1, 1]> input_13 = mul(x = gate_states_1, y = up_states_1)[name = string("input_13")];
            string hidden_states_7_pad_type_0 = const()[name = string("hidden_states_7_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_7_strides_0 = const()[name = string("hidden_states_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_7_pad_0 = const()[name = string("hidden_states_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_7_dilations_0 = const()[name = string("hidden_states_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_7_groups_0 = const()[name = string("hidden_states_7_groups_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 1, 1]> hidden_states_7 = conv(dilations = hidden_states_7_dilations_0, groups = hidden_states_7_groups_0, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = hidden_states_7_strides_0, weight = model_model_layers_0_mlp_down_proj_weight_palettized, x = input_13)[name = string("hidden_states_7")];
            tensor<int32, [1]> var_447_axes_0 = const()[name = string("op_447_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 1]> var_447 = squeeze(axes = var_447_axes_0, x = hidden_states_7)[name = string("op_447")];
            tensor<int32, [3]> var_448 = const()[name = string("op_448"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 1, 2048]> var_449 = transpose(perm = var_448, x = var_447)[name = string("transpose_28")];
            tensor<fp16, [1, 1, 2048]> hidden_states_9_cast_fp16 = add(x = hidden_states_5_cast_fp16, y = var_449)[name = string("hidden_states_9_cast_fp16")];
            tensor<int32, [1]> mean_5_axes_0 = const()[name = string("mean_5_axes_0"), val = tensor<int32, [1]>([-1])];
            bool mean_5_keep_dims_0 = const()[name = string("mean_5_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 1, 1]> mean_5_cast_fp16 = reduce_mean(axes = mean_5_axes_0, keep_dims = mean_5_keep_dims_0, x = hidden_states_9_cast_fp16)[name = string("mean_5_cast_fp16")];
            tensor<fp16, [1, 1, 2048]> input_15_cast_fp16 = sub(x = hidden_states_9_cast_fp16, y = mean_5_cast_fp16)[name = string("input_15_cast_fp16")];
            tensor<int32, [1]> var_457_axes_0 = const()[name = string("op_457_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2048]> model_model_layers_1_input_layernorm_weight_to_fp16 = const()[name = string("model_model_layers_1_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262859264)))];
            tensor<fp16, [1, 1, 2048]> var_457_cast_fp16 = layer_norm(axes = var_457_axes_0, epsilon = var_44_to_fp16, gamma = model_model_layers_1_input_layernorm_weight_to_fp16, x = input_15_cast_fp16)[name = string("op_457_cast_fp16")];
            tensor<int32, [3]> var_460 = const()[name = string("op_460"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_462_axes_0 = const()[name = string("op_462_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 1]> var_461 = transpose(perm = var_460, x = var_457_cast_fp16)[name = string("transpose_27")];
            tensor<fp16, [1, 2048, 1, 1]> var_462 = expand_dims(axes = var_462_axes_0, x = var_461)[name = string("op_462")];
            string var_469_pad_type_0 = const()[name = string("op_469_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_469_strides_0 = const()[name = string("op_469_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_469_pad_0 = const()[name = string("op_469_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_469_dilations_0 = const()[name = string("op_469_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_469_groups_0 = const()[name = string("op_469_groups_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 1, 1]> var_469 = conv(dilations = var_469_dilations_0, groups = var_469_groups_0, pad = var_469_pad_0, pad_type = var_469_pad_type_0, strides = var_469_strides_0, weight = model_model_layers_1_self_attn_q_proj_weight_palettized, x = var_462)[name = string("op_469")];
            tensor<int32, [4]> var_470 = const()[name = string("op_470"), val = tensor<int32, [4]>([1, 32, 1, 64])];
            tensor<fp16, [1, 32, 1, 64]> var_471 = reshape(shape = var_470, x = var_469)[name = string("op_471")];
            string var_478_pad_type_0 = const()[name = string("op_478_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_478_strides_0 = const()[name = string("op_478_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_478_pad_0 = const()[name = string("op_478_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_478_dilations_0 = const()[name = string("op_478_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_478_groups_0 = const()[name = string("op_478_groups_0"), val = int32(1)];
            tensor<fp16, [1, 512, 1, 1]> var_478 = conv(dilations = var_478_dilations_0, groups = var_478_groups_0, pad = var_478_pad_0, pad_type = var_478_pad_type_0, strides = var_478_strides_0, weight = model_model_layers_1_self_attn_k_proj_weight_palettized, x = var_462)[name = string("op_478")];
            tensor<int32, [4]> var_479 = const()[name = string("op_479"), val = tensor<int32, [4]>([1, 8, 1, 64])];
            tensor<fp16, [1, 8, 1, 64]> var_480 = reshape(shape = var_479, x = var_478)[name = string("op_480")];
            string var_487_pad_type_0 = const()[name = string("op_487_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_487_strides_0 = const()[name = string("op_487_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_487_pad_0 = const()[name = string("op_487_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_487_dilations_0 = const()[name = string("op_487_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_487_groups_0 = const()[name = string("op_487_groups_0"), val = int32(1)];
            tensor<fp16, [1, 512, 1, 1]> var_487 = conv(dilations = var_487_dilations_0, groups = var_487_groups_0, pad = var_487_pad_0, pad_type = var_487_pad_type_0, strides = var_487_strides_0, weight = model_model_layers_1_self_attn_v_proj_weight_palettized, x = var_462)[name = string("op_487")];
            tensor<int32, [4]> var_488 = const()[name = string("op_488"), val = tensor<int32, [4]>([1, 8, 1, 64])];
            tensor<fp16, [1, 8, 1, 64]> var_489 = reshape(shape = var_488, x = var_487)[name = string("op_489")];
            tensor<int32, [4]> x1_5_begin_0 = const()[name = string("x1_5_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_5_end_0 = const()[name = string("x1_5_end_0"), val = tensor<int32, [4]>([1, 32, 1, 32])];
            tensor<bool, [4]> x1_5_end_mask_0 = const()[name = string("x1_5_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 1, 32]> x1_5 = slice_by_index(begin = x1_5_begin_0, end = x1_5_end_0, end_mask = x1_5_end_mask_0, x = var_471)[name = string("x1_5")];
            tensor<int32, [4]> x2_5_begin_0 = const()[name = string("x2_5_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_5_end_0 = const()[name = string("x2_5_end_0"), val = tensor<int32, [4]>([1, 32, 1, 64])];
            tensor<bool, [4]> x2_5_end_mask_0 = const()[name = string("x2_5_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 1, 32]> x2_5 = slice_by_index(begin = x2_5_begin_0, end = x2_5_end_0, end_mask = x2_5_end_mask_0, x = var_471)[name = string("x2_5")];
            tensor<fp16, [1, 32, 1, 32]> var_503_cast_fp16 = mul(x = x1_5, y = cos_3_cast_fp16)[name = string("op_503_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_504_cast_fp16 = mul(x = x2_5, y = sin_3_cast_fp16)[name = string("op_504_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_505_cast_fp16 = sub(x = var_503_cast_fp16, y = var_504_cast_fp16)[name = string("op_505_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_506_cast_fp16 = mul(x = x2_5, y = cos_3_cast_fp16)[name = string("op_506_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_507_cast_fp16 = mul(x = x1_5, y = sin_3_cast_fp16)[name = string("op_507_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_508_cast_fp16 = add(x = var_506_cast_fp16, y = var_507_cast_fp16)[name = string("op_508_cast_fp16")];
            bool rotated_5_interleave_0 = const()[name = string("rotated_5_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 32, 1, 64]> rotated_5_cast_fp16 = concat(axis = var_49, interleave = rotated_5_interleave_0, values = (var_505_cast_fp16, var_508_cast_fp16))[name = string("rotated_5_cast_fp16")];
            tensor<int32, [4]> x1_7_begin_0 = const()[name = string("x1_7_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_7_end_0 = const()[name = string("x1_7_end_0"), val = tensor<int32, [4]>([1, 8, 1, 32])];
            tensor<bool, [4]> x1_7_end_mask_0 = const()[name = string("x1_7_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 1, 32]> x1_7 = slice_by_index(begin = x1_7_begin_0, end = x1_7_end_0, end_mask = x1_7_end_mask_0, x = var_480)[name = string("x1_7")];
            tensor<int32, [4]> x2_7_begin_0 = const()[name = string("x2_7_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_7_end_0 = const()[name = string("x2_7_end_0"), val = tensor<int32, [4]>([1, 8, 1, 64])];
            tensor<bool, [4]> x2_7_end_mask_0 = const()[name = string("x2_7_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 1, 32]> x2_7 = slice_by_index(begin = x2_7_begin_0, end = x2_7_end_0, end_mask = x2_7_end_mask_0, x = var_480)[name = string("x2_7")];
            tensor<fp16, [1, 8, 1, 32]> var_524_cast_fp16 = mul(x = x1_7, y = cos_3_cast_fp16)[name = string("op_524_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_525_cast_fp16 = mul(x = x2_7, y = sin_3_cast_fp16)[name = string("op_525_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_526_cast_fp16 = sub(x = var_524_cast_fp16, y = var_525_cast_fp16)[name = string("op_526_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_527_cast_fp16 = mul(x = x2_7, y = cos_3_cast_fp16)[name = string("op_527_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_528_cast_fp16 = mul(x = x1_7, y = sin_3_cast_fp16)[name = string("op_528_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_529_cast_fp16 = add(x = var_527_cast_fp16, y = var_528_cast_fp16)[name = string("op_529_cast_fp16")];
            bool rotated_7_interleave_0 = const()[name = string("rotated_7_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 8, 1, 64]> rotated_7_cast_fp16 = concat(axis = var_49, interleave = rotated_7_interleave_0, values = (var_526_cast_fp16, var_529_cast_fp16))[name = string("rotated_7_cast_fp16")];
            tensor<int32, [1]> expand_dims_12 = const()[name = string("expand_dims_12"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> expand_dims_13 = const()[name = string("expand_dims_13"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_15 = const()[name = string("expand_dims_15"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_16 = const()[name = string("expand_dims_16"), val = tensor<int32, [1]>([2])];
            int32 concat_10_axis_0 = const()[name = string("concat_10_axis_0"), val = int32(0)];
            bool concat_10_interleave_0 = const()[name = string("concat_10_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_10 = concat(axis = concat_10_axis_0, interleave = concat_10_interleave_0, values = (expand_dims_12, expand_dims_13, current_pos, expand_dims_15))[name = string("concat_10")];
            tensor<int32, [1]> concat_11_values1_0 = const()[name = string("concat_11_values1_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> concat_11_values3_0 = const()[name = string("concat_11_values3_0"), val = tensor<int32, [1]>([0])];
            int32 concat_11_axis_0 = const()[name = string("concat_11_axis_0"), val = int32(0)];
            bool concat_11_interleave_0 = const()[name = string("concat_11_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_11 = concat(axis = concat_11_axis_0, interleave = concat_11_interleave_0, values = (expand_dims_16, concat_11_values1_0, var_349, concat_11_values3_0))[name = string("concat_11")];
            tensor<int32, [4]> model_model_kv_cache_0_internal_tensor_assign_3_stride_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_3_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_3_begin_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_3_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_3_end_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_3_end_mask_0"), val = tensor<bool, [4]>([false, true, false, true])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_3_squeeze_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_3_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<fp16, [32, 8, 2048, 64]> model_model_kv_cache_0_internal_tensor_assign_3_cast_fp16 = slice_update(begin = concat_10, begin_mask = model_model_kv_cache_0_internal_tensor_assign_3_begin_mask_0, end = concat_11, end_mask = model_model_kv_cache_0_internal_tensor_assign_3_end_mask_0, squeeze_mask = model_model_kv_cache_0_internal_tensor_assign_3_squeeze_mask_0, stride = model_model_kv_cache_0_internal_tensor_assign_3_stride_0, update = rotated_7_cast_fp16, x = coreml_update_state_17)[name = string("model_model_kv_cache_0_internal_tensor_assign_3_cast_fp16")];
            write_state(data = model_model_kv_cache_0_internal_tensor_assign_3_cast_fp16, input = model_model_kv_cache_0)[name = string("coreml_update_state_2_write_state")];
            tensor<fp16, [32, 8, 2048, 64]> coreml_update_state_18 = read_state(input = model_model_kv_cache_0)[name = string("coreml_update_state_2")];
            tensor<int32, [1]> expand_dims_18 = const()[name = string("expand_dims_18"), val = tensor<int32, [1]>([17])];
            tensor<int32, [1]> expand_dims_19 = const()[name = string("expand_dims_19"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_21 = const()[name = string("expand_dims_21"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_22 = const()[name = string("expand_dims_22"), val = tensor<int32, [1]>([18])];
            int32 concat_14_axis_0 = const()[name = string("concat_14_axis_0"), val = int32(0)];
            bool concat_14_interleave_0 = const()[name = string("concat_14_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_14 = concat(axis = concat_14_axis_0, interleave = concat_14_interleave_0, values = (expand_dims_18, expand_dims_19, current_pos, expand_dims_21))[name = string("concat_14")];
            tensor<int32, [1]> concat_15_values1_0 = const()[name = string("concat_15_values1_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> concat_15_values3_0 = const()[name = string("concat_15_values3_0"), val = tensor<int32, [1]>([0])];
            int32 concat_15_axis_0 = const()[name = string("concat_15_axis_0"), val = int32(0)];
            bool concat_15_interleave_0 = const()[name = string("concat_15_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_15 = concat(axis = concat_15_axis_0, interleave = concat_15_interleave_0, values = (expand_dims_22, concat_15_values1_0, var_349, concat_15_values3_0))[name = string("concat_15")];
            tensor<int32, [4]> model_model_kv_cache_0_internal_tensor_assign_4_stride_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_4_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_4_begin_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_4_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_4_end_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_4_end_mask_0"), val = tensor<bool, [4]>([false, true, false, true])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_4_squeeze_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_4_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<fp16, [32, 8, 2048, 64]> model_model_kv_cache_0_internal_tensor_assign_4_cast_fp16 = slice_update(begin = concat_14, begin_mask = model_model_kv_cache_0_internal_tensor_assign_4_begin_mask_0, end = concat_15, end_mask = model_model_kv_cache_0_internal_tensor_assign_4_end_mask_0, squeeze_mask = model_model_kv_cache_0_internal_tensor_assign_4_squeeze_mask_0, stride = model_model_kv_cache_0_internal_tensor_assign_4_stride_0, update = var_489, x = coreml_update_state_18)[name = string("model_model_kv_cache_0_internal_tensor_assign_4_cast_fp16")];
            write_state(data = model_model_kv_cache_0_internal_tensor_assign_4_cast_fp16, input = model_model_kv_cache_0)[name = string("coreml_update_state_3_write_state")];
            tensor<fp16, [32, 8, 2048, 64]> coreml_update_state_19 = read_state(input = model_model_kv_cache_0)[name = string("coreml_update_state_3")];
            tensor<int32, [4]> var_549_begin_0 = const()[name = string("op_549_begin_0"), val = tensor<int32, [4]>([1, 0, 0, 0])];
            tensor<int32, [4]> var_549_end_0 = const()[name = string("op_549_end_0"), val = tensor<int32, [4]>([2, 8, 2048, 64])];
            tensor<bool, [4]> var_549_end_mask_0 = const()[name = string("op_549_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<fp16, [1, 8, 2048, 64]> var_549_cast_fp16 = slice_by_index(begin = var_549_begin_0, end = var_549_end_0, end_mask = var_549_end_mask_0, x = coreml_update_state_19)[name = string("op_549_cast_fp16")];
            tensor<int32, [1]> K_layer_cache_3_axes_0 = const()[name = string("K_layer_cache_3_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<fp16, [8, 2048, 64]> K_layer_cache_3_cast_fp16 = squeeze(axes = K_layer_cache_3_axes_0, x = var_549_cast_fp16)[name = string("K_layer_cache_3_cast_fp16")];
            tensor<int32, [4]> var_551_begin_0 = const()[name = string("op_551_begin_0"), val = tensor<int32, [4]>([17, 0, 0, 0])];
            tensor<int32, [4]> var_551_end_0 = const()[name = string("op_551_end_0"), val = tensor<int32, [4]>([18, 8, 2048, 64])];
            tensor<bool, [4]> var_551_end_mask_0 = const()[name = string("op_551_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<fp16, [1, 8, 2048, 64]> var_551_cast_fp16 = slice_by_index(begin = var_551_begin_0, end = var_551_end_0, end_mask = var_551_end_mask_0, x = coreml_update_state_19)[name = string("op_551_cast_fp16")];
            tensor<int32, [1]> V_layer_cache_3_axes_0 = const()[name = string("V_layer_cache_3_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<fp16, [8, 2048, 64]> V_layer_cache_3_cast_fp16 = squeeze(axes = V_layer_cache_3_axes_0, x = var_551_cast_fp16)[name = string("V_layer_cache_3_cast_fp16")];
            tensor<int32, [1]> x_39_axes_0 = const()[name = string("x_39_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [8, 1, 2048, 64]> x_39_cast_fp16 = expand_dims(axes = x_39_axes_0, x = K_layer_cache_3_cast_fp16)[name = string("x_39_cast_fp16")];
            tensor<int32, [4]> var_560 = const()[name = string("op_560"), val = tensor<int32, [4]>([1, 4, 1, 1])];
            tensor<fp16, [8, 4, 2048, 64]> x_41_cast_fp16 = tile(reps = var_560, x = x_39_cast_fp16)[name = string("x_41_cast_fp16")];
            tensor<int32, [4]> var_564 = const()[name = string("op_564"), val = tensor<int32, [4]>([1, -1, 2048, 64])];
            tensor<fp16, [1, 32, 2048, 64]> key_states_7_cast_fp16 = reshape(shape = var_564, x = x_41_cast_fp16)[name = string("key_states_7_cast_fp16")];
            tensor<int32, [1]> x_45_axes_0 = const()[name = string("x_45_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [8, 1, 2048, 64]> x_45_cast_fp16 = expand_dims(axes = x_45_axes_0, x = V_layer_cache_3_cast_fp16)[name = string("x_45_cast_fp16")];
            tensor<int32, [4]> var_567 = const()[name = string("op_567"), val = tensor<int32, [4]>([1, 4, 1, 1])];
            tensor<fp16, [8, 4, 2048, 64]> x_47_cast_fp16 = tile(reps = var_567, x = x_45_cast_fp16)[name = string("x_47_cast_fp16")];
            tensor<int32, [4]> var_571 = const()[name = string("op_571"), val = tensor<int32, [4]>([1, -1, 2048, 64])];
            tensor<fp16, [1, 32, 2048, 64]> value_states_7_cast_fp16 = reshape(shape = var_571, x = x_47_cast_fp16)[name = string("value_states_7_cast_fp16")];
            bool var_574_transpose_x_1 = const()[name = string("op_574_transpose_x_1"), val = bool(false)];
            bool var_574_transpose_y_1 = const()[name = string("op_574_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 2048]> var_574_cast_fp16 = matmul(transpose_x = var_574_transpose_x_1, transpose_y = var_574_transpose_y_1, x = rotated_5_cast_fp16, y = key_states_7_cast_fp16)[name = string("op_574_cast_fp16")];
            fp16 var_575_to_fp16 = const()[name = string("op_575_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 32, 1, 2048]> attn_weights_5_cast_fp16 = mul(x = var_574_cast_fp16, y = var_575_to_fp16)[name = string("attn_weights_5_cast_fp16")];
            tensor<fp16, [1, 32, 1, 2048]> x_49_cast_fp16 = add(x = attn_weights_5_cast_fp16, y = causal_mask)[name = string("x_49_cast_fp16")];
            tensor<int32, [1]> reduce_max_1_axes_0 = const()[name = string("reduce_max_1_axes_0"), val = tensor<int32, [1]>([-1])];
            bool reduce_max_1_keep_dims_0 = const()[name = string("reduce_max_1_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1]> reduce_max_1_cast_fp16 = reduce_max(axes = reduce_max_1_axes_0, keep_dims = reduce_max_1_keep_dims_0, x = x_49_cast_fp16)[name = string("reduce_max_1_cast_fp16")];
            tensor<fp16, [1, 32, 1, 2048]> x_51_cast_fp16 = sub(x = x_49_cast_fp16, y = reduce_max_1_cast_fp16)[name = string("x_51_cast_fp16")];
            tensor<fp16, [1, 32, 1, 2048]> exp_x_3_cast_fp16 = exp(x = x_51_cast_fp16)[name = string("exp_x_3_cast_fp16")];
            tensor<int32, [1]> var_586_axes_0 = const()[name = string("op_586_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_586_keep_dims_0 = const()[name = string("op_586_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1]> var_586_cast_fp16 = reduce_sum(axes = var_586_axes_0, keep_dims = var_586_keep_dims_0, x = exp_x_3_cast_fp16)[name = string("op_586_cast_fp16")];
            tensor<fp16, [1, 32, 1, 2048]> attn_weights_7_cast_fp16 = real_div(x = exp_x_3_cast_fp16, y = var_586_cast_fp16)[name = string("attn_weights_7_cast_fp16")];
            bool attn_output_7_transpose_x_0 = const()[name = string("attn_output_7_transpose_x_0"), val = bool(false)];
            bool attn_output_7_transpose_y_0 = const()[name = string("attn_output_7_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 32, 1, 64]> attn_output_7_cast_fp16 = matmul(transpose_x = attn_output_7_transpose_x_0, transpose_y = attn_output_7_transpose_y_0, x = attn_weights_7_cast_fp16, y = value_states_7_cast_fp16)[name = string("attn_output_7_cast_fp16")];
            tensor<int32, [4]> var_589_perm_0 = const()[name = string("op_589_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_591 = const()[name = string("op_591"), val = tensor<int32, [3]>([1, 1, 2048])];
            tensor<fp16, [1, 1, 32, 64]> var_589_cast_fp16 = transpose(perm = var_589_perm_0, x = attn_output_7_cast_fp16)[name = string("transpose_26")];
            tensor<fp16, [1, 1, 2048]> input_19_cast_fp16 = reshape(shape = var_591, x = var_589_cast_fp16)[name = string("input_19_cast_fp16")];
            tensor<fp16, [2048, 2048]> model_model_layers_1_self_attn_o_proj_weight_promoted_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262863424))), lut = tensor<fp16, [256, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(264960640))))[name = string("model_model_layers_1_self_attn_o_proj_weight_promoted_to_fp16_palettized")];
            tensor<fp16, [1, 1, 2048]> linear_1_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_1_self_attn_o_proj_weight_promoted_to_fp16_palettized, x = input_19_cast_fp16)[name = string("linear_1_cast_fp16")];
            tensor<fp16, [1, 1, 2048]> hidden_states_13_cast_fp16 = add(x = hidden_states_9_cast_fp16, y = linear_1_cast_fp16)[name = string("hidden_states_13_cast_fp16")];
            tensor<int32, [1]> mean_7_axes_0 = const()[name = string("mean_7_axes_0"), val = tensor<int32, [1]>([-1])];
            bool mean_7_keep_dims_0 = const()[name = string("mean_7_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 1, 1]> mean_7_cast_fp16 = reduce_mean(axes = mean_7_axes_0, keep_dims = mean_7_keep_dims_0, x = hidden_states_13_cast_fp16)[name = string("mean_7_cast_fp16")];
            tensor<fp16, [1, 1, 2048]> input_21_cast_fp16 = sub(x = hidden_states_13_cast_fp16, y = mean_7_cast_fp16)[name = string("input_21_cast_fp16")];
            tensor<int32, [1]> var_602_axes_0 = const()[name = string("op_602_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2048]> model_model_layers_1_post_attention_layernorm_weight_to_fp16 = const()[name = string("model_model_layers_1_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(264968896)))];
            tensor<fp16, [1, 1, 2048]> var_602_cast_fp16 = layer_norm(axes = var_602_axes_0, epsilon = var_44_to_fp16, gamma = model_model_layers_1_post_attention_layernorm_weight_to_fp16, x = input_21_cast_fp16)[name = string("op_602_cast_fp16")];
            tensor<int32, [3]> var_609 = const()[name = string("op_609"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> input_23_axes_0 = const()[name = string("input_23_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 1]> var_610 = transpose(perm = var_609, x = var_602_cast_fp16)[name = string("transpose_25")];
            tensor<fp16, [1, 2048, 1, 1]> input_23 = expand_dims(axes = input_23_axes_0, x = var_610)[name = string("input_23")];
            string input_25_pad_type_0 = const()[name = string("input_25_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_25_strides_0 = const()[name = string("input_25_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_25_pad_0 = const()[name = string("input_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_25_dilations_0 = const()[name = string("input_25_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_25_groups_0 = const()[name = string("input_25_groups_0"), val = int32(1)];
            tensor<fp16, [1, 8192, 1, 1]> input_25 = conv(dilations = input_25_dilations_0, groups = input_25_groups_0, pad = input_25_pad_0, pad_type = input_25_pad_type_0, strides = input_25_strides_0, weight = model_model_layers_1_mlp_gate_proj_weight_palettized, x = input_23)[name = string("input_25")];
            string up_states_3_pad_type_0 = const()[name = string("up_states_3_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> up_states_3_strides_0 = const()[name = string("up_states_3_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> up_states_3_pad_0 = const()[name = string("up_states_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> up_states_3_dilations_0 = const()[name = string("up_states_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 up_states_3_groups_0 = const()[name = string("up_states_3_groups_0"), val = int32(1)];
            tensor<fp16, [1, 8192, 1, 1]> up_states_3 = conv(dilations = up_states_3_dilations_0, groups = up_states_3_groups_0, pad = up_states_3_pad_0, pad_type = up_states_3_pad_type_0, strides = up_states_3_strides_0, weight = model_model_layers_1_mlp_up_proj_weight_palettized, x = input_23)[name = string("up_states_3")];
            tensor<fp16, [1, 8192, 1, 1]> gate_states_3 = silu(x = input_25)[name = string("gate_states_3")];
            tensor<fp16, [1, 8192, 1, 1]> input_27 = mul(x = gate_states_3, y = up_states_3)[name = string("input_27")];
            string hidden_states_15_pad_type_0 = const()[name = string("hidden_states_15_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_15_strides_0 = const()[name = string("hidden_states_15_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_15_pad_0 = const()[name = string("hidden_states_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_15_dilations_0 = const()[name = string("hidden_states_15_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_15_groups_0 = const()[name = string("hidden_states_15_groups_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 1, 1]> hidden_states_15 = conv(dilations = hidden_states_15_dilations_0, groups = hidden_states_15_groups_0, pad = hidden_states_15_pad_0, pad_type = hidden_states_15_pad_type_0, strides = hidden_states_15_strides_0, weight = model_model_layers_1_mlp_down_proj_weight_palettized, x = input_27)[name = string("hidden_states_15")];
            tensor<int32, [1]> var_632_axes_0 = const()[name = string("op_632_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 1]> var_632 = squeeze(axes = var_632_axes_0, x = hidden_states_15)[name = string("op_632")];
            tensor<int32, [3]> var_633 = const()[name = string("op_633"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 1, 2048]> var_634 = transpose(perm = var_633, x = var_632)[name = string("transpose_24")];
            tensor<fp16, [1, 1, 2048]> hidden_states_17_cast_fp16 = add(x = hidden_states_13_cast_fp16, y = var_634)[name = string("hidden_states_17_cast_fp16")];
            tensor<int32, [1]> mean_9_axes_0 = const()[name = string("mean_9_axes_0"), val = tensor<int32, [1]>([-1])];
            bool mean_9_keep_dims_0 = const()[name = string("mean_9_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 1, 1]> mean_9_cast_fp16 = reduce_mean(axes = mean_9_axes_0, keep_dims = mean_9_keep_dims_0, x = hidden_states_17_cast_fp16)[name = string("mean_9_cast_fp16")];
            tensor<fp16, [1, 1, 2048]> input_29_cast_fp16 = sub(x = hidden_states_17_cast_fp16, y = mean_9_cast_fp16)[name = string("input_29_cast_fp16")];
            tensor<int32, [1]> var_642_axes_0 = const()[name = string("op_642_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2048]> model_model_layers_2_input_layernorm_weight_to_fp16 = const()[name = string("model_model_layers_2_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(264973056)))];
            tensor<fp16, [1, 1, 2048]> var_642_cast_fp16 = layer_norm(axes = var_642_axes_0, epsilon = var_44_to_fp16, gamma = model_model_layers_2_input_layernorm_weight_to_fp16, x = input_29_cast_fp16)[name = string("op_642_cast_fp16")];
            tensor<int32, [3]> var_645 = const()[name = string("op_645"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_647_axes_0 = const()[name = string("op_647_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 1]> var_646 = transpose(perm = var_645, x = var_642_cast_fp16)[name = string("transpose_23")];
            tensor<fp16, [1, 2048, 1, 1]> var_647 = expand_dims(axes = var_647_axes_0, x = var_646)[name = string("op_647")];
            string var_654_pad_type_0 = const()[name = string("op_654_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_654_strides_0 = const()[name = string("op_654_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_654_pad_0 = const()[name = string("op_654_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_654_dilations_0 = const()[name = string("op_654_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_654_groups_0 = const()[name = string("op_654_groups_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 1, 1]> var_654 = conv(dilations = var_654_dilations_0, groups = var_654_groups_0, pad = var_654_pad_0, pad_type = var_654_pad_type_0, strides = var_654_strides_0, weight = model_model_layers_2_self_attn_q_proj_weight_palettized, x = var_647)[name = string("op_654")];
            tensor<int32, [4]> var_655 = const()[name = string("op_655"), val = tensor<int32, [4]>([1, 32, 1, 64])];
            tensor<fp16, [1, 32, 1, 64]> var_656 = reshape(shape = var_655, x = var_654)[name = string("op_656")];
            string var_663_pad_type_0 = const()[name = string("op_663_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_663_strides_0 = const()[name = string("op_663_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_663_pad_0 = const()[name = string("op_663_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_663_dilations_0 = const()[name = string("op_663_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_663_groups_0 = const()[name = string("op_663_groups_0"), val = int32(1)];
            tensor<fp16, [1, 512, 1, 1]> var_663 = conv(dilations = var_663_dilations_0, groups = var_663_groups_0, pad = var_663_pad_0, pad_type = var_663_pad_type_0, strides = var_663_strides_0, weight = model_model_layers_2_self_attn_k_proj_weight_palettized, x = var_647)[name = string("op_663")];
            tensor<int32, [4]> var_664 = const()[name = string("op_664"), val = tensor<int32, [4]>([1, 8, 1, 64])];
            tensor<fp16, [1, 8, 1, 64]> var_665 = reshape(shape = var_664, x = var_663)[name = string("op_665")];
            string var_672_pad_type_0 = const()[name = string("op_672_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_672_strides_0 = const()[name = string("op_672_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_672_pad_0 = const()[name = string("op_672_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_672_dilations_0 = const()[name = string("op_672_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_672_groups_0 = const()[name = string("op_672_groups_0"), val = int32(1)];
            tensor<fp16, [1, 512, 1, 1]> var_672 = conv(dilations = var_672_dilations_0, groups = var_672_groups_0, pad = var_672_pad_0, pad_type = var_672_pad_type_0, strides = var_672_strides_0, weight = model_model_layers_2_self_attn_v_proj_weight_palettized, x = var_647)[name = string("op_672")];
            tensor<int32, [4]> var_673 = const()[name = string("op_673"), val = tensor<int32, [4]>([1, 8, 1, 64])];
            tensor<fp16, [1, 8, 1, 64]> var_674 = reshape(shape = var_673, x = var_672)[name = string("op_674")];
            tensor<int32, [4]> x1_9_begin_0 = const()[name = string("x1_9_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_9_end_0 = const()[name = string("x1_9_end_0"), val = tensor<int32, [4]>([1, 32, 1, 32])];
            tensor<bool, [4]> x1_9_end_mask_0 = const()[name = string("x1_9_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 1, 32]> x1_9 = slice_by_index(begin = x1_9_begin_0, end = x1_9_end_0, end_mask = x1_9_end_mask_0, x = var_656)[name = string("x1_9")];
            tensor<int32, [4]> x2_9_begin_0 = const()[name = string("x2_9_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_9_end_0 = const()[name = string("x2_9_end_0"), val = tensor<int32, [4]>([1, 32, 1, 64])];
            tensor<bool, [4]> x2_9_end_mask_0 = const()[name = string("x2_9_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 1, 32]> x2_9 = slice_by_index(begin = x2_9_begin_0, end = x2_9_end_0, end_mask = x2_9_end_mask_0, x = var_656)[name = string("x2_9")];
            tensor<fp16, [1, 32, 1, 32]> var_688_cast_fp16 = mul(x = x1_9, y = cos_3_cast_fp16)[name = string("op_688_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_689_cast_fp16 = mul(x = x2_9, y = sin_3_cast_fp16)[name = string("op_689_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_690_cast_fp16 = sub(x = var_688_cast_fp16, y = var_689_cast_fp16)[name = string("op_690_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_691_cast_fp16 = mul(x = x2_9, y = cos_3_cast_fp16)[name = string("op_691_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_692_cast_fp16 = mul(x = x1_9, y = sin_3_cast_fp16)[name = string("op_692_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_693_cast_fp16 = add(x = var_691_cast_fp16, y = var_692_cast_fp16)[name = string("op_693_cast_fp16")];
            bool rotated_9_interleave_0 = const()[name = string("rotated_9_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 32, 1, 64]> rotated_9_cast_fp16 = concat(axis = var_49, interleave = rotated_9_interleave_0, values = (var_690_cast_fp16, var_693_cast_fp16))[name = string("rotated_9_cast_fp16")];
            tensor<int32, [4]> x1_11_begin_0 = const()[name = string("x1_11_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_11_end_0 = const()[name = string("x1_11_end_0"), val = tensor<int32, [4]>([1, 8, 1, 32])];
            tensor<bool, [4]> x1_11_end_mask_0 = const()[name = string("x1_11_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 1, 32]> x1_11 = slice_by_index(begin = x1_11_begin_0, end = x1_11_end_0, end_mask = x1_11_end_mask_0, x = var_665)[name = string("x1_11")];
            tensor<int32, [4]> x2_11_begin_0 = const()[name = string("x2_11_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_11_end_0 = const()[name = string("x2_11_end_0"), val = tensor<int32, [4]>([1, 8, 1, 64])];
            tensor<bool, [4]> x2_11_end_mask_0 = const()[name = string("x2_11_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 1, 32]> x2_11 = slice_by_index(begin = x2_11_begin_0, end = x2_11_end_0, end_mask = x2_11_end_mask_0, x = var_665)[name = string("x2_11")];
            tensor<fp16, [1, 8, 1, 32]> var_709_cast_fp16 = mul(x = x1_11, y = cos_3_cast_fp16)[name = string("op_709_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_710_cast_fp16 = mul(x = x2_11, y = sin_3_cast_fp16)[name = string("op_710_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_711_cast_fp16 = sub(x = var_709_cast_fp16, y = var_710_cast_fp16)[name = string("op_711_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_712_cast_fp16 = mul(x = x2_11, y = cos_3_cast_fp16)[name = string("op_712_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_713_cast_fp16 = mul(x = x1_11, y = sin_3_cast_fp16)[name = string("op_713_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_714_cast_fp16 = add(x = var_712_cast_fp16, y = var_713_cast_fp16)[name = string("op_714_cast_fp16")];
            bool rotated_11_interleave_0 = const()[name = string("rotated_11_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 8, 1, 64]> rotated_11_cast_fp16 = concat(axis = var_49, interleave = rotated_11_interleave_0, values = (var_711_cast_fp16, var_714_cast_fp16))[name = string("rotated_11_cast_fp16")];
            tensor<int32, [1]> expand_dims_24 = const()[name = string("expand_dims_24"), val = tensor<int32, [1]>([2])];
            tensor<int32, [1]> expand_dims_25 = const()[name = string("expand_dims_25"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_27 = const()[name = string("expand_dims_27"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_28 = const()[name = string("expand_dims_28"), val = tensor<int32, [1]>([3])];
            int32 concat_18_axis_0 = const()[name = string("concat_18_axis_0"), val = int32(0)];
            bool concat_18_interleave_0 = const()[name = string("concat_18_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_18 = concat(axis = concat_18_axis_0, interleave = concat_18_interleave_0, values = (expand_dims_24, expand_dims_25, current_pos, expand_dims_27))[name = string("concat_18")];
            tensor<int32, [1]> concat_19_values1_0 = const()[name = string("concat_19_values1_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> concat_19_values3_0 = const()[name = string("concat_19_values3_0"), val = tensor<int32, [1]>([0])];
            int32 concat_19_axis_0 = const()[name = string("concat_19_axis_0"), val = int32(0)];
            bool concat_19_interleave_0 = const()[name = string("concat_19_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_19 = concat(axis = concat_19_axis_0, interleave = concat_19_interleave_0, values = (expand_dims_28, concat_19_values1_0, var_349, concat_19_values3_0))[name = string("concat_19")];
            tensor<int32, [4]> model_model_kv_cache_0_internal_tensor_assign_5_stride_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_5_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_5_begin_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_5_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_5_end_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_5_end_mask_0"), val = tensor<bool, [4]>([false, true, false, true])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_5_squeeze_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_5_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<fp16, [32, 8, 2048, 64]> model_model_kv_cache_0_internal_tensor_assign_5_cast_fp16 = slice_update(begin = concat_18, begin_mask = model_model_kv_cache_0_internal_tensor_assign_5_begin_mask_0, end = concat_19, end_mask = model_model_kv_cache_0_internal_tensor_assign_5_end_mask_0, squeeze_mask = model_model_kv_cache_0_internal_tensor_assign_5_squeeze_mask_0, stride = model_model_kv_cache_0_internal_tensor_assign_5_stride_0, update = rotated_11_cast_fp16, x = coreml_update_state_19)[name = string("model_model_kv_cache_0_internal_tensor_assign_5_cast_fp16")];
            write_state(data = model_model_kv_cache_0_internal_tensor_assign_5_cast_fp16, input = model_model_kv_cache_0)[name = string("coreml_update_state_4_write_state")];
            tensor<fp16, [32, 8, 2048, 64]> coreml_update_state_20 = read_state(input = model_model_kv_cache_0)[name = string("coreml_update_state_4")];
            tensor<int32, [1]> expand_dims_30 = const()[name = string("expand_dims_30"), val = tensor<int32, [1]>([18])];
            tensor<int32, [1]> expand_dims_31 = const()[name = string("expand_dims_31"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_33 = const()[name = string("expand_dims_33"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_34 = const()[name = string("expand_dims_34"), val = tensor<int32, [1]>([19])];
            int32 concat_22_axis_0 = const()[name = string("concat_22_axis_0"), val = int32(0)];
            bool concat_22_interleave_0 = const()[name = string("concat_22_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_22 = concat(axis = concat_22_axis_0, interleave = concat_22_interleave_0, values = (expand_dims_30, expand_dims_31, current_pos, expand_dims_33))[name = string("concat_22")];
            tensor<int32, [1]> concat_23_values1_0 = const()[name = string("concat_23_values1_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> concat_23_values3_0 = const()[name = string("concat_23_values3_0"), val = tensor<int32, [1]>([0])];
            int32 concat_23_axis_0 = const()[name = string("concat_23_axis_0"), val = int32(0)];
            bool concat_23_interleave_0 = const()[name = string("concat_23_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_23 = concat(axis = concat_23_axis_0, interleave = concat_23_interleave_0, values = (expand_dims_34, concat_23_values1_0, var_349, concat_23_values3_0))[name = string("concat_23")];
            tensor<int32, [4]> model_model_kv_cache_0_internal_tensor_assign_6_stride_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_6_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_6_begin_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_6_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_6_end_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_6_end_mask_0"), val = tensor<bool, [4]>([false, true, false, true])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_6_squeeze_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_6_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<fp16, [32, 8, 2048, 64]> model_model_kv_cache_0_internal_tensor_assign_6_cast_fp16 = slice_update(begin = concat_22, begin_mask = model_model_kv_cache_0_internal_tensor_assign_6_begin_mask_0, end = concat_23, end_mask = model_model_kv_cache_0_internal_tensor_assign_6_end_mask_0, squeeze_mask = model_model_kv_cache_0_internal_tensor_assign_6_squeeze_mask_0, stride = model_model_kv_cache_0_internal_tensor_assign_6_stride_0, update = var_674, x = coreml_update_state_20)[name = string("model_model_kv_cache_0_internal_tensor_assign_6_cast_fp16")];
            write_state(data = model_model_kv_cache_0_internal_tensor_assign_6_cast_fp16, input = model_model_kv_cache_0)[name = string("coreml_update_state_5_write_state")];
            tensor<fp16, [32, 8, 2048, 64]> coreml_update_state_21 = read_state(input = model_model_kv_cache_0)[name = string("coreml_update_state_5")];
            tensor<int32, [4]> var_734_begin_0 = const()[name = string("op_734_begin_0"), val = tensor<int32, [4]>([2, 0, 0, 0])];
            tensor<int32, [4]> var_734_end_0 = const()[name = string("op_734_end_0"), val = tensor<int32, [4]>([3, 8, 2048, 64])];
            tensor<bool, [4]> var_734_end_mask_0 = const()[name = string("op_734_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<fp16, [1, 8, 2048, 64]> var_734_cast_fp16 = slice_by_index(begin = var_734_begin_0, end = var_734_end_0, end_mask = var_734_end_mask_0, x = coreml_update_state_21)[name = string("op_734_cast_fp16")];
            tensor<int32, [1]> K_layer_cache_5_axes_0 = const()[name = string("K_layer_cache_5_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<fp16, [8, 2048, 64]> K_layer_cache_5_cast_fp16 = squeeze(axes = K_layer_cache_5_axes_0, x = var_734_cast_fp16)[name = string("K_layer_cache_5_cast_fp16")];
            tensor<int32, [4]> var_736_begin_0 = const()[name = string("op_736_begin_0"), val = tensor<int32, [4]>([18, 0, 0, 0])];
            tensor<int32, [4]> var_736_end_0 = const()[name = string("op_736_end_0"), val = tensor<int32, [4]>([19, 8, 2048, 64])];
            tensor<bool, [4]> var_736_end_mask_0 = const()[name = string("op_736_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<fp16, [1, 8, 2048, 64]> var_736_cast_fp16 = slice_by_index(begin = var_736_begin_0, end = var_736_end_0, end_mask = var_736_end_mask_0, x = coreml_update_state_21)[name = string("op_736_cast_fp16")];
            tensor<int32, [1]> V_layer_cache_5_axes_0 = const()[name = string("V_layer_cache_5_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<fp16, [8, 2048, 64]> V_layer_cache_5_cast_fp16 = squeeze(axes = V_layer_cache_5_axes_0, x = var_736_cast_fp16)[name = string("V_layer_cache_5_cast_fp16")];
            tensor<int32, [1]> x_67_axes_0 = const()[name = string("x_67_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [8, 1, 2048, 64]> x_67_cast_fp16 = expand_dims(axes = x_67_axes_0, x = K_layer_cache_5_cast_fp16)[name = string("x_67_cast_fp16")];
            tensor<int32, [4]> var_745 = const()[name = string("op_745"), val = tensor<int32, [4]>([1, 4, 1, 1])];
            tensor<fp16, [8, 4, 2048, 64]> x_69_cast_fp16 = tile(reps = var_745, x = x_67_cast_fp16)[name = string("x_69_cast_fp16")];
            tensor<int32, [4]> var_749 = const()[name = string("op_749"), val = tensor<int32, [4]>([1, -1, 2048, 64])];
            tensor<fp16, [1, 32, 2048, 64]> key_states_11_cast_fp16 = reshape(shape = var_749, x = x_69_cast_fp16)[name = string("key_states_11_cast_fp16")];
            tensor<int32, [1]> x_73_axes_0 = const()[name = string("x_73_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [8, 1, 2048, 64]> x_73_cast_fp16 = expand_dims(axes = x_73_axes_0, x = V_layer_cache_5_cast_fp16)[name = string("x_73_cast_fp16")];
            tensor<int32, [4]> var_752 = const()[name = string("op_752"), val = tensor<int32, [4]>([1, 4, 1, 1])];
            tensor<fp16, [8, 4, 2048, 64]> x_75_cast_fp16 = tile(reps = var_752, x = x_73_cast_fp16)[name = string("x_75_cast_fp16")];
            tensor<int32, [4]> var_756 = const()[name = string("op_756"), val = tensor<int32, [4]>([1, -1, 2048, 64])];
            tensor<fp16, [1, 32, 2048, 64]> value_states_11_cast_fp16 = reshape(shape = var_756, x = x_75_cast_fp16)[name = string("value_states_11_cast_fp16")];
            bool var_759_transpose_x_1 = const()[name = string("op_759_transpose_x_1"), val = bool(false)];
            bool var_759_transpose_y_1 = const()[name = string("op_759_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 2048]> var_759_cast_fp16 = matmul(transpose_x = var_759_transpose_x_1, transpose_y = var_759_transpose_y_1, x = rotated_9_cast_fp16, y = key_states_11_cast_fp16)[name = string("op_759_cast_fp16")];
            fp16 var_760_to_fp16 = const()[name = string("op_760_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 32, 1, 2048]> attn_weights_9_cast_fp16 = mul(x = var_759_cast_fp16, y = var_760_to_fp16)[name = string("attn_weights_9_cast_fp16")];
            tensor<fp16, [1, 32, 1, 2048]> x_77_cast_fp16 = add(x = attn_weights_9_cast_fp16, y = causal_mask)[name = string("x_77_cast_fp16")];
            tensor<int32, [1]> reduce_max_2_axes_0 = const()[name = string("reduce_max_2_axes_0"), val = tensor<int32, [1]>([-1])];
            bool reduce_max_2_keep_dims_0 = const()[name = string("reduce_max_2_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1]> reduce_max_2_cast_fp16 = reduce_max(axes = reduce_max_2_axes_0, keep_dims = reduce_max_2_keep_dims_0, x = x_77_cast_fp16)[name = string("reduce_max_2_cast_fp16")];
            tensor<fp16, [1, 32, 1, 2048]> x_79_cast_fp16 = sub(x = x_77_cast_fp16, y = reduce_max_2_cast_fp16)[name = string("x_79_cast_fp16")];
            tensor<fp16, [1, 32, 1, 2048]> exp_x_5_cast_fp16 = exp(x = x_79_cast_fp16)[name = string("exp_x_5_cast_fp16")];
            tensor<int32, [1]> var_771_axes_0 = const()[name = string("op_771_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_771_keep_dims_0 = const()[name = string("op_771_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1]> var_771_cast_fp16 = reduce_sum(axes = var_771_axes_0, keep_dims = var_771_keep_dims_0, x = exp_x_5_cast_fp16)[name = string("op_771_cast_fp16")];
            tensor<fp16, [1, 32, 1, 2048]> attn_weights_11_cast_fp16 = real_div(x = exp_x_5_cast_fp16, y = var_771_cast_fp16)[name = string("attn_weights_11_cast_fp16")];
            bool attn_output_13_transpose_x_0 = const()[name = string("attn_output_13_transpose_x_0"), val = bool(false)];
            bool attn_output_13_transpose_y_0 = const()[name = string("attn_output_13_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 32, 1, 64]> attn_output_13_cast_fp16 = matmul(transpose_x = attn_output_13_transpose_x_0, transpose_y = attn_output_13_transpose_y_0, x = attn_weights_11_cast_fp16, y = value_states_11_cast_fp16)[name = string("attn_output_13_cast_fp16")];
            tensor<int32, [4]> var_774_perm_0 = const()[name = string("op_774_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_776 = const()[name = string("op_776"), val = tensor<int32, [3]>([1, 1, 2048])];
            tensor<fp16, [1, 1, 32, 64]> var_774_cast_fp16 = transpose(perm = var_774_perm_0, x = attn_output_13_cast_fp16)[name = string("transpose_22")];
            tensor<fp16, [1, 1, 2048]> input_33_cast_fp16 = reshape(shape = var_776, x = var_774_cast_fp16)[name = string("input_33_cast_fp16")];
            tensor<fp16, [2048, 2048]> model_model_layers_2_self_attn_o_proj_weight_promoted_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(264977216))), lut = tensor<fp16, [256, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267074432))))[name = string("model_model_layers_2_self_attn_o_proj_weight_promoted_to_fp16_palettized")];
            tensor<fp16, [1, 1, 2048]> linear_2_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_2_self_attn_o_proj_weight_promoted_to_fp16_palettized, x = input_33_cast_fp16)[name = string("linear_2_cast_fp16")];
            tensor<fp16, [1, 1, 2048]> hidden_states_21_cast_fp16 = add(x = hidden_states_17_cast_fp16, y = linear_2_cast_fp16)[name = string("hidden_states_21_cast_fp16")];
            tensor<int32, [1]> mean_11_axes_0 = const()[name = string("mean_11_axes_0"), val = tensor<int32, [1]>([-1])];
            bool mean_11_keep_dims_0 = const()[name = string("mean_11_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 1, 1]> mean_11_cast_fp16 = reduce_mean(axes = mean_11_axes_0, keep_dims = mean_11_keep_dims_0, x = hidden_states_21_cast_fp16)[name = string("mean_11_cast_fp16")];
            tensor<fp16, [1, 1, 2048]> input_35_cast_fp16 = sub(x = hidden_states_21_cast_fp16, y = mean_11_cast_fp16)[name = string("input_35_cast_fp16")];
            tensor<int32, [1]> var_787_axes_0 = const()[name = string("op_787_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2048]> model_model_layers_2_post_attention_layernorm_weight_to_fp16 = const()[name = string("model_model_layers_2_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267082688)))];
            tensor<fp16, [1, 1, 2048]> var_787_cast_fp16 = layer_norm(axes = var_787_axes_0, epsilon = var_44_to_fp16, gamma = model_model_layers_2_post_attention_layernorm_weight_to_fp16, x = input_35_cast_fp16)[name = string("op_787_cast_fp16")];
            tensor<int32, [3]> var_794 = const()[name = string("op_794"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> input_37_axes_0 = const()[name = string("input_37_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 1]> var_795 = transpose(perm = var_794, x = var_787_cast_fp16)[name = string("transpose_21")];
            tensor<fp16, [1, 2048, 1, 1]> input_37 = expand_dims(axes = input_37_axes_0, x = var_795)[name = string("input_37")];
            string input_39_pad_type_0 = const()[name = string("input_39_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_39_strides_0 = const()[name = string("input_39_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_39_pad_0 = const()[name = string("input_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_39_dilations_0 = const()[name = string("input_39_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_39_groups_0 = const()[name = string("input_39_groups_0"), val = int32(1)];
            tensor<fp16, [1, 8192, 1, 1]> input_39 = conv(dilations = input_39_dilations_0, groups = input_39_groups_0, pad = input_39_pad_0, pad_type = input_39_pad_type_0, strides = input_39_strides_0, weight = model_model_layers_2_mlp_gate_proj_weight_palettized, x = input_37)[name = string("input_39")];
            string up_states_5_pad_type_0 = const()[name = string("up_states_5_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> up_states_5_strides_0 = const()[name = string("up_states_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> up_states_5_pad_0 = const()[name = string("up_states_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> up_states_5_dilations_0 = const()[name = string("up_states_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 up_states_5_groups_0 = const()[name = string("up_states_5_groups_0"), val = int32(1)];
            tensor<fp16, [1, 8192, 1, 1]> up_states_5 = conv(dilations = up_states_5_dilations_0, groups = up_states_5_groups_0, pad = up_states_5_pad_0, pad_type = up_states_5_pad_type_0, strides = up_states_5_strides_0, weight = model_model_layers_2_mlp_up_proj_weight_palettized, x = input_37)[name = string("up_states_5")];
            tensor<fp16, [1, 8192, 1, 1]> gate_states_5 = silu(x = input_39)[name = string("gate_states_5")];
            tensor<fp16, [1, 8192, 1, 1]> input_41 = mul(x = gate_states_5, y = up_states_5)[name = string("input_41")];
            string hidden_states_23_pad_type_0 = const()[name = string("hidden_states_23_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_23_strides_0 = const()[name = string("hidden_states_23_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_23_pad_0 = const()[name = string("hidden_states_23_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_23_dilations_0 = const()[name = string("hidden_states_23_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_23_groups_0 = const()[name = string("hidden_states_23_groups_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 1, 1]> hidden_states_23 = conv(dilations = hidden_states_23_dilations_0, groups = hidden_states_23_groups_0, pad = hidden_states_23_pad_0, pad_type = hidden_states_23_pad_type_0, strides = hidden_states_23_strides_0, weight = model_model_layers_2_mlp_down_proj_weight_palettized, x = input_41)[name = string("hidden_states_23")];
            tensor<int32, [1]> var_817_axes_0 = const()[name = string("op_817_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 1]> var_817 = squeeze(axes = var_817_axes_0, x = hidden_states_23)[name = string("op_817")];
            tensor<int32, [3]> var_818 = const()[name = string("op_818"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 1, 2048]> var_819 = transpose(perm = var_818, x = var_817)[name = string("transpose_20")];
            tensor<fp16, [1, 1, 2048]> hidden_states_25_cast_fp16 = add(x = hidden_states_21_cast_fp16, y = var_819)[name = string("hidden_states_25_cast_fp16")];
            tensor<int32, [1]> mean_13_axes_0 = const()[name = string("mean_13_axes_0"), val = tensor<int32, [1]>([-1])];
            bool mean_13_keep_dims_0 = const()[name = string("mean_13_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 1, 1]> mean_13_cast_fp16 = reduce_mean(axes = mean_13_axes_0, keep_dims = mean_13_keep_dims_0, x = hidden_states_25_cast_fp16)[name = string("mean_13_cast_fp16")];
            tensor<fp16, [1, 1, 2048]> input_43_cast_fp16 = sub(x = hidden_states_25_cast_fp16, y = mean_13_cast_fp16)[name = string("input_43_cast_fp16")];
            tensor<int32, [1]> var_827_axes_0 = const()[name = string("op_827_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2048]> model_model_layers_3_input_layernorm_weight_to_fp16 = const()[name = string("model_model_layers_3_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267086848)))];
            tensor<fp16, [1, 1, 2048]> var_827_cast_fp16 = layer_norm(axes = var_827_axes_0, epsilon = var_44_to_fp16, gamma = model_model_layers_3_input_layernorm_weight_to_fp16, x = input_43_cast_fp16)[name = string("op_827_cast_fp16")];
            tensor<int32, [3]> var_830 = const()[name = string("op_830"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_832_axes_0 = const()[name = string("op_832_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 1]> var_831 = transpose(perm = var_830, x = var_827_cast_fp16)[name = string("transpose_19")];
            tensor<fp16, [1, 2048, 1, 1]> var_832 = expand_dims(axes = var_832_axes_0, x = var_831)[name = string("op_832")];
            string var_839_pad_type_0 = const()[name = string("op_839_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_839_strides_0 = const()[name = string("op_839_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_839_pad_0 = const()[name = string("op_839_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_839_dilations_0 = const()[name = string("op_839_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_839_groups_0 = const()[name = string("op_839_groups_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 1, 1]> var_839 = conv(dilations = var_839_dilations_0, groups = var_839_groups_0, pad = var_839_pad_0, pad_type = var_839_pad_type_0, strides = var_839_strides_0, weight = model_model_layers_3_self_attn_q_proj_weight_palettized, x = var_832)[name = string("op_839")];
            tensor<int32, [4]> var_840 = const()[name = string("op_840"), val = tensor<int32, [4]>([1, 32, 1, 64])];
            tensor<fp16, [1, 32, 1, 64]> var_841 = reshape(shape = var_840, x = var_839)[name = string("op_841")];
            string var_848_pad_type_0 = const()[name = string("op_848_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_848_strides_0 = const()[name = string("op_848_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_848_pad_0 = const()[name = string("op_848_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_848_dilations_0 = const()[name = string("op_848_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_848_groups_0 = const()[name = string("op_848_groups_0"), val = int32(1)];
            tensor<fp16, [1, 512, 1, 1]> var_848 = conv(dilations = var_848_dilations_0, groups = var_848_groups_0, pad = var_848_pad_0, pad_type = var_848_pad_type_0, strides = var_848_strides_0, weight = model_model_layers_3_self_attn_k_proj_weight_palettized, x = var_832)[name = string("op_848")];
            tensor<int32, [4]> var_849 = const()[name = string("op_849"), val = tensor<int32, [4]>([1, 8, 1, 64])];
            tensor<fp16, [1, 8, 1, 64]> var_850 = reshape(shape = var_849, x = var_848)[name = string("op_850")];
            string var_857_pad_type_0 = const()[name = string("op_857_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_857_strides_0 = const()[name = string("op_857_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_857_pad_0 = const()[name = string("op_857_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_857_dilations_0 = const()[name = string("op_857_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_857_groups_0 = const()[name = string("op_857_groups_0"), val = int32(1)];
            tensor<fp16, [1, 512, 1, 1]> var_857 = conv(dilations = var_857_dilations_0, groups = var_857_groups_0, pad = var_857_pad_0, pad_type = var_857_pad_type_0, strides = var_857_strides_0, weight = model_model_layers_3_self_attn_v_proj_weight_palettized, x = var_832)[name = string("op_857")];
            tensor<int32, [4]> var_858 = const()[name = string("op_858"), val = tensor<int32, [4]>([1, 8, 1, 64])];
            tensor<fp16, [1, 8, 1, 64]> var_859 = reshape(shape = var_858, x = var_857)[name = string("op_859")];
            tensor<int32, [4]> x1_13_begin_0 = const()[name = string("x1_13_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_13_end_0 = const()[name = string("x1_13_end_0"), val = tensor<int32, [4]>([1, 32, 1, 32])];
            tensor<bool, [4]> x1_13_end_mask_0 = const()[name = string("x1_13_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 1, 32]> x1_13 = slice_by_index(begin = x1_13_begin_0, end = x1_13_end_0, end_mask = x1_13_end_mask_0, x = var_841)[name = string("x1_13")];
            tensor<int32, [4]> x2_13_begin_0 = const()[name = string("x2_13_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_13_end_0 = const()[name = string("x2_13_end_0"), val = tensor<int32, [4]>([1, 32, 1, 64])];
            tensor<bool, [4]> x2_13_end_mask_0 = const()[name = string("x2_13_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 1, 32]> x2_13 = slice_by_index(begin = x2_13_begin_0, end = x2_13_end_0, end_mask = x2_13_end_mask_0, x = var_841)[name = string("x2_13")];
            tensor<fp16, [1, 32, 1, 32]> var_873_cast_fp16 = mul(x = x1_13, y = cos_3_cast_fp16)[name = string("op_873_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_874_cast_fp16 = mul(x = x2_13, y = sin_3_cast_fp16)[name = string("op_874_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_875_cast_fp16 = sub(x = var_873_cast_fp16, y = var_874_cast_fp16)[name = string("op_875_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_876_cast_fp16 = mul(x = x2_13, y = cos_3_cast_fp16)[name = string("op_876_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_877_cast_fp16 = mul(x = x1_13, y = sin_3_cast_fp16)[name = string("op_877_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_878_cast_fp16 = add(x = var_876_cast_fp16, y = var_877_cast_fp16)[name = string("op_878_cast_fp16")];
            bool rotated_13_interleave_0 = const()[name = string("rotated_13_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 32, 1, 64]> rotated_13_cast_fp16 = concat(axis = var_49, interleave = rotated_13_interleave_0, values = (var_875_cast_fp16, var_878_cast_fp16))[name = string("rotated_13_cast_fp16")];
            tensor<int32, [4]> x1_15_begin_0 = const()[name = string("x1_15_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_15_end_0 = const()[name = string("x1_15_end_0"), val = tensor<int32, [4]>([1, 8, 1, 32])];
            tensor<bool, [4]> x1_15_end_mask_0 = const()[name = string("x1_15_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 1, 32]> x1_15 = slice_by_index(begin = x1_15_begin_0, end = x1_15_end_0, end_mask = x1_15_end_mask_0, x = var_850)[name = string("x1_15")];
            tensor<int32, [4]> x2_15_begin_0 = const()[name = string("x2_15_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_15_end_0 = const()[name = string("x2_15_end_0"), val = tensor<int32, [4]>([1, 8, 1, 64])];
            tensor<bool, [4]> x2_15_end_mask_0 = const()[name = string("x2_15_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 1, 32]> x2_15 = slice_by_index(begin = x2_15_begin_0, end = x2_15_end_0, end_mask = x2_15_end_mask_0, x = var_850)[name = string("x2_15")];
            tensor<fp16, [1, 8, 1, 32]> var_894_cast_fp16 = mul(x = x1_15, y = cos_3_cast_fp16)[name = string("op_894_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_895_cast_fp16 = mul(x = x2_15, y = sin_3_cast_fp16)[name = string("op_895_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_896_cast_fp16 = sub(x = var_894_cast_fp16, y = var_895_cast_fp16)[name = string("op_896_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_897_cast_fp16 = mul(x = x2_15, y = cos_3_cast_fp16)[name = string("op_897_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_898_cast_fp16 = mul(x = x1_15, y = sin_3_cast_fp16)[name = string("op_898_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_899_cast_fp16 = add(x = var_897_cast_fp16, y = var_898_cast_fp16)[name = string("op_899_cast_fp16")];
            bool rotated_15_interleave_0 = const()[name = string("rotated_15_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 8, 1, 64]> rotated_15_cast_fp16 = concat(axis = var_49, interleave = rotated_15_interleave_0, values = (var_896_cast_fp16, var_899_cast_fp16))[name = string("rotated_15_cast_fp16")];
            tensor<int32, [1]> expand_dims_36 = const()[name = string("expand_dims_36"), val = tensor<int32, [1]>([3])];
            tensor<int32, [1]> expand_dims_37 = const()[name = string("expand_dims_37"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_39 = const()[name = string("expand_dims_39"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_40 = const()[name = string("expand_dims_40"), val = tensor<int32, [1]>([4])];
            int32 concat_26_axis_0 = const()[name = string("concat_26_axis_0"), val = int32(0)];
            bool concat_26_interleave_0 = const()[name = string("concat_26_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_26 = concat(axis = concat_26_axis_0, interleave = concat_26_interleave_0, values = (expand_dims_36, expand_dims_37, current_pos, expand_dims_39))[name = string("concat_26")];
            tensor<int32, [1]> concat_27_values1_0 = const()[name = string("concat_27_values1_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> concat_27_values3_0 = const()[name = string("concat_27_values3_0"), val = tensor<int32, [1]>([0])];
            int32 concat_27_axis_0 = const()[name = string("concat_27_axis_0"), val = int32(0)];
            bool concat_27_interleave_0 = const()[name = string("concat_27_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_27 = concat(axis = concat_27_axis_0, interleave = concat_27_interleave_0, values = (expand_dims_40, concat_27_values1_0, var_349, concat_27_values3_0))[name = string("concat_27")];
            tensor<int32, [4]> model_model_kv_cache_0_internal_tensor_assign_7_stride_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_7_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_7_begin_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_7_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_7_end_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_7_end_mask_0"), val = tensor<bool, [4]>([false, true, false, true])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_7_squeeze_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_7_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<fp16, [32, 8, 2048, 64]> model_model_kv_cache_0_internal_tensor_assign_7_cast_fp16 = slice_update(begin = concat_26, begin_mask = model_model_kv_cache_0_internal_tensor_assign_7_begin_mask_0, end = concat_27, end_mask = model_model_kv_cache_0_internal_tensor_assign_7_end_mask_0, squeeze_mask = model_model_kv_cache_0_internal_tensor_assign_7_squeeze_mask_0, stride = model_model_kv_cache_0_internal_tensor_assign_7_stride_0, update = rotated_15_cast_fp16, x = coreml_update_state_21)[name = string("model_model_kv_cache_0_internal_tensor_assign_7_cast_fp16")];
            write_state(data = model_model_kv_cache_0_internal_tensor_assign_7_cast_fp16, input = model_model_kv_cache_0)[name = string("coreml_update_state_6_write_state")];
            tensor<fp16, [32, 8, 2048, 64]> coreml_update_state_22 = read_state(input = model_model_kv_cache_0)[name = string("coreml_update_state_6")];
            tensor<int32, [1]> expand_dims_42 = const()[name = string("expand_dims_42"), val = tensor<int32, [1]>([19])];
            tensor<int32, [1]> expand_dims_43 = const()[name = string("expand_dims_43"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_45 = const()[name = string("expand_dims_45"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_46 = const()[name = string("expand_dims_46"), val = tensor<int32, [1]>([20])];
            int32 concat_30_axis_0 = const()[name = string("concat_30_axis_0"), val = int32(0)];
            bool concat_30_interleave_0 = const()[name = string("concat_30_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_30 = concat(axis = concat_30_axis_0, interleave = concat_30_interleave_0, values = (expand_dims_42, expand_dims_43, current_pos, expand_dims_45))[name = string("concat_30")];
            tensor<int32, [1]> concat_31_values1_0 = const()[name = string("concat_31_values1_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> concat_31_values3_0 = const()[name = string("concat_31_values3_0"), val = tensor<int32, [1]>([0])];
            int32 concat_31_axis_0 = const()[name = string("concat_31_axis_0"), val = int32(0)];
            bool concat_31_interleave_0 = const()[name = string("concat_31_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_31 = concat(axis = concat_31_axis_0, interleave = concat_31_interleave_0, values = (expand_dims_46, concat_31_values1_0, var_349, concat_31_values3_0))[name = string("concat_31")];
            tensor<int32, [4]> model_model_kv_cache_0_internal_tensor_assign_8_stride_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_8_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_8_begin_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_8_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_8_end_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_8_end_mask_0"), val = tensor<bool, [4]>([false, true, false, true])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_8_squeeze_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_8_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<fp16, [32, 8, 2048, 64]> model_model_kv_cache_0_internal_tensor_assign_8_cast_fp16 = slice_update(begin = concat_30, begin_mask = model_model_kv_cache_0_internal_tensor_assign_8_begin_mask_0, end = concat_31, end_mask = model_model_kv_cache_0_internal_tensor_assign_8_end_mask_0, squeeze_mask = model_model_kv_cache_0_internal_tensor_assign_8_squeeze_mask_0, stride = model_model_kv_cache_0_internal_tensor_assign_8_stride_0, update = var_859, x = coreml_update_state_22)[name = string("model_model_kv_cache_0_internal_tensor_assign_8_cast_fp16")];
            write_state(data = model_model_kv_cache_0_internal_tensor_assign_8_cast_fp16, input = model_model_kv_cache_0)[name = string("coreml_update_state_7_write_state")];
            tensor<fp16, [32, 8, 2048, 64]> coreml_update_state_23 = read_state(input = model_model_kv_cache_0)[name = string("coreml_update_state_7")];
            tensor<int32, [4]> var_919_begin_0 = const()[name = string("op_919_begin_0"), val = tensor<int32, [4]>([3, 0, 0, 0])];
            tensor<int32, [4]> var_919_end_0 = const()[name = string("op_919_end_0"), val = tensor<int32, [4]>([4, 8, 2048, 64])];
            tensor<bool, [4]> var_919_end_mask_0 = const()[name = string("op_919_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<fp16, [1, 8, 2048, 64]> var_919_cast_fp16 = slice_by_index(begin = var_919_begin_0, end = var_919_end_0, end_mask = var_919_end_mask_0, x = coreml_update_state_23)[name = string("op_919_cast_fp16")];
            tensor<int32, [1]> K_layer_cache_7_axes_0 = const()[name = string("K_layer_cache_7_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<fp16, [8, 2048, 64]> K_layer_cache_7_cast_fp16 = squeeze(axes = K_layer_cache_7_axes_0, x = var_919_cast_fp16)[name = string("K_layer_cache_7_cast_fp16")];
            tensor<int32, [4]> var_921_begin_0 = const()[name = string("op_921_begin_0"), val = tensor<int32, [4]>([19, 0, 0, 0])];
            tensor<int32, [4]> var_921_end_0 = const()[name = string("op_921_end_0"), val = tensor<int32, [4]>([20, 8, 2048, 64])];
            tensor<bool, [4]> var_921_end_mask_0 = const()[name = string("op_921_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<fp16, [1, 8, 2048, 64]> var_921_cast_fp16 = slice_by_index(begin = var_921_begin_0, end = var_921_end_0, end_mask = var_921_end_mask_0, x = coreml_update_state_23)[name = string("op_921_cast_fp16")];
            tensor<int32, [1]> V_layer_cache_7_axes_0 = const()[name = string("V_layer_cache_7_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<fp16, [8, 2048, 64]> V_layer_cache_7_cast_fp16 = squeeze(axes = V_layer_cache_7_axes_0, x = var_921_cast_fp16)[name = string("V_layer_cache_7_cast_fp16")];
            tensor<int32, [1]> x_95_axes_0 = const()[name = string("x_95_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [8, 1, 2048, 64]> x_95_cast_fp16 = expand_dims(axes = x_95_axes_0, x = K_layer_cache_7_cast_fp16)[name = string("x_95_cast_fp16")];
            tensor<int32, [4]> var_930 = const()[name = string("op_930"), val = tensor<int32, [4]>([1, 4, 1, 1])];
            tensor<fp16, [8, 4, 2048, 64]> x_97_cast_fp16 = tile(reps = var_930, x = x_95_cast_fp16)[name = string("x_97_cast_fp16")];
            tensor<int32, [4]> var_934 = const()[name = string("op_934"), val = tensor<int32, [4]>([1, -1, 2048, 64])];
            tensor<fp16, [1, 32, 2048, 64]> key_states_15_cast_fp16 = reshape(shape = var_934, x = x_97_cast_fp16)[name = string("key_states_15_cast_fp16")];
            tensor<int32, [1]> x_101_axes_0 = const()[name = string("x_101_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [8, 1, 2048, 64]> x_101_cast_fp16 = expand_dims(axes = x_101_axes_0, x = V_layer_cache_7_cast_fp16)[name = string("x_101_cast_fp16")];
            tensor<int32, [4]> var_937 = const()[name = string("op_937"), val = tensor<int32, [4]>([1, 4, 1, 1])];
            tensor<fp16, [8, 4, 2048, 64]> x_103_cast_fp16 = tile(reps = var_937, x = x_101_cast_fp16)[name = string("x_103_cast_fp16")];
            tensor<int32, [4]> var_941 = const()[name = string("op_941"), val = tensor<int32, [4]>([1, -1, 2048, 64])];
            tensor<fp16, [1, 32, 2048, 64]> value_states_15_cast_fp16 = reshape(shape = var_941, x = x_103_cast_fp16)[name = string("value_states_15_cast_fp16")];
            bool var_944_transpose_x_1 = const()[name = string("op_944_transpose_x_1"), val = bool(false)];
            bool var_944_transpose_y_1 = const()[name = string("op_944_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 2048]> var_944_cast_fp16 = matmul(transpose_x = var_944_transpose_x_1, transpose_y = var_944_transpose_y_1, x = rotated_13_cast_fp16, y = key_states_15_cast_fp16)[name = string("op_944_cast_fp16")];
            fp16 var_945_to_fp16 = const()[name = string("op_945_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 32, 1, 2048]> attn_weights_13_cast_fp16 = mul(x = var_944_cast_fp16, y = var_945_to_fp16)[name = string("attn_weights_13_cast_fp16")];
            tensor<fp16, [1, 32, 1, 2048]> x_105_cast_fp16 = add(x = attn_weights_13_cast_fp16, y = causal_mask)[name = string("x_105_cast_fp16")];
            tensor<int32, [1]> reduce_max_3_axes_0 = const()[name = string("reduce_max_3_axes_0"), val = tensor<int32, [1]>([-1])];
            bool reduce_max_3_keep_dims_0 = const()[name = string("reduce_max_3_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1]> reduce_max_3_cast_fp16 = reduce_max(axes = reduce_max_3_axes_0, keep_dims = reduce_max_3_keep_dims_0, x = x_105_cast_fp16)[name = string("reduce_max_3_cast_fp16")];
            tensor<fp16, [1, 32, 1, 2048]> x_107_cast_fp16 = sub(x = x_105_cast_fp16, y = reduce_max_3_cast_fp16)[name = string("x_107_cast_fp16")];
            tensor<fp16, [1, 32, 1, 2048]> exp_x_7_cast_fp16 = exp(x = x_107_cast_fp16)[name = string("exp_x_7_cast_fp16")];
            tensor<int32, [1]> var_956_axes_0 = const()[name = string("op_956_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_956_keep_dims_0 = const()[name = string("op_956_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1]> var_956_cast_fp16 = reduce_sum(axes = var_956_axes_0, keep_dims = var_956_keep_dims_0, x = exp_x_7_cast_fp16)[name = string("op_956_cast_fp16")];
            tensor<fp16, [1, 32, 1, 2048]> attn_weights_15_cast_fp16 = real_div(x = exp_x_7_cast_fp16, y = var_956_cast_fp16)[name = string("attn_weights_15_cast_fp16")];
            bool attn_output_19_transpose_x_0 = const()[name = string("attn_output_19_transpose_x_0"), val = bool(false)];
            bool attn_output_19_transpose_y_0 = const()[name = string("attn_output_19_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 32, 1, 64]> attn_output_19_cast_fp16 = matmul(transpose_x = attn_output_19_transpose_x_0, transpose_y = attn_output_19_transpose_y_0, x = attn_weights_15_cast_fp16, y = value_states_15_cast_fp16)[name = string("attn_output_19_cast_fp16")];
            tensor<int32, [4]> var_959_perm_0 = const()[name = string("op_959_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_961 = const()[name = string("op_961"), val = tensor<int32, [3]>([1, 1, 2048])];
            tensor<fp16, [1, 1, 32, 64]> var_959_cast_fp16 = transpose(perm = var_959_perm_0, x = attn_output_19_cast_fp16)[name = string("transpose_18")];
            tensor<fp16, [1, 1, 2048]> input_47_cast_fp16 = reshape(shape = var_961, x = var_959_cast_fp16)[name = string("input_47_cast_fp16")];
            tensor<fp16, [2048, 2048]> model_model_layers_3_self_attn_o_proj_weight_promoted_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267091008))), lut = tensor<fp16, [256, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269188224))))[name = string("model_model_layers_3_self_attn_o_proj_weight_promoted_to_fp16_palettized")];
            tensor<fp16, [1, 1, 2048]> linear_3_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_3_self_attn_o_proj_weight_promoted_to_fp16_palettized, x = input_47_cast_fp16)[name = string("linear_3_cast_fp16")];
            tensor<fp16, [1, 1, 2048]> hidden_states_29_cast_fp16 = add(x = hidden_states_25_cast_fp16, y = linear_3_cast_fp16)[name = string("hidden_states_29_cast_fp16")];
            tensor<int32, [1]> mean_15_axes_0 = const()[name = string("mean_15_axes_0"), val = tensor<int32, [1]>([-1])];
            bool mean_15_keep_dims_0 = const()[name = string("mean_15_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 1, 1]> mean_15_cast_fp16 = reduce_mean(axes = mean_15_axes_0, keep_dims = mean_15_keep_dims_0, x = hidden_states_29_cast_fp16)[name = string("mean_15_cast_fp16")];
            tensor<fp16, [1, 1, 2048]> input_49_cast_fp16 = sub(x = hidden_states_29_cast_fp16, y = mean_15_cast_fp16)[name = string("input_49_cast_fp16")];
            tensor<int32, [1]> var_972_axes_0 = const()[name = string("op_972_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2048]> model_model_layers_3_post_attention_layernorm_weight_to_fp16 = const()[name = string("model_model_layers_3_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269196480)))];
            tensor<fp16, [1, 1, 2048]> var_972_cast_fp16 = layer_norm(axes = var_972_axes_0, epsilon = var_44_to_fp16, gamma = model_model_layers_3_post_attention_layernorm_weight_to_fp16, x = input_49_cast_fp16)[name = string("op_972_cast_fp16")];
            tensor<int32, [3]> var_979 = const()[name = string("op_979"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> input_51_axes_0 = const()[name = string("input_51_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 1]> var_980 = transpose(perm = var_979, x = var_972_cast_fp16)[name = string("transpose_17")];
            tensor<fp16, [1, 2048, 1, 1]> input_51 = expand_dims(axes = input_51_axes_0, x = var_980)[name = string("input_51")];
            string input_53_pad_type_0 = const()[name = string("input_53_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_53_strides_0 = const()[name = string("input_53_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_53_pad_0 = const()[name = string("input_53_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_53_dilations_0 = const()[name = string("input_53_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_53_groups_0 = const()[name = string("input_53_groups_0"), val = int32(1)];
            tensor<fp16, [1, 8192, 1, 1]> input_53 = conv(dilations = input_53_dilations_0, groups = input_53_groups_0, pad = input_53_pad_0, pad_type = input_53_pad_type_0, strides = input_53_strides_0, weight = model_model_layers_3_mlp_gate_proj_weight_palettized, x = input_51)[name = string("input_53")];
            string up_states_7_pad_type_0 = const()[name = string("up_states_7_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> up_states_7_strides_0 = const()[name = string("up_states_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> up_states_7_pad_0 = const()[name = string("up_states_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> up_states_7_dilations_0 = const()[name = string("up_states_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 up_states_7_groups_0 = const()[name = string("up_states_7_groups_0"), val = int32(1)];
            tensor<fp16, [1, 8192, 1, 1]> up_states_7 = conv(dilations = up_states_7_dilations_0, groups = up_states_7_groups_0, pad = up_states_7_pad_0, pad_type = up_states_7_pad_type_0, strides = up_states_7_strides_0, weight = model_model_layers_3_mlp_up_proj_weight_palettized, x = input_51)[name = string("up_states_7")];
            tensor<fp16, [1, 8192, 1, 1]> gate_states_7 = silu(x = input_53)[name = string("gate_states_7")];
            tensor<fp16, [1, 8192, 1, 1]> input_55 = mul(x = gate_states_7, y = up_states_7)[name = string("input_55")];
            string hidden_states_31_pad_type_0 = const()[name = string("hidden_states_31_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_31_strides_0 = const()[name = string("hidden_states_31_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_31_pad_0 = const()[name = string("hidden_states_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_31_dilations_0 = const()[name = string("hidden_states_31_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_31_groups_0 = const()[name = string("hidden_states_31_groups_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 1, 1]> hidden_states_31 = conv(dilations = hidden_states_31_dilations_0, groups = hidden_states_31_groups_0, pad = hidden_states_31_pad_0, pad_type = hidden_states_31_pad_type_0, strides = hidden_states_31_strides_0, weight = model_model_layers_3_mlp_down_proj_weight_palettized, x = input_55)[name = string("hidden_states_31")];
            tensor<int32, [1]> var_1002_axes_0 = const()[name = string("op_1002_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 1]> var_1002 = squeeze(axes = var_1002_axes_0, x = hidden_states_31)[name = string("op_1002")];
            tensor<int32, [3]> var_1003 = const()[name = string("op_1003"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 1, 2048]> var_1004 = transpose(perm = var_1003, x = var_1002)[name = string("transpose_16")];
            tensor<fp16, [1, 1, 2048]> hidden_states_33_cast_fp16 = add(x = hidden_states_29_cast_fp16, y = var_1004)[name = string("hidden_states_33_cast_fp16")];
            tensor<int32, [1]> mean_17_axes_0 = const()[name = string("mean_17_axes_0"), val = tensor<int32, [1]>([-1])];
            bool mean_17_keep_dims_0 = const()[name = string("mean_17_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 1, 1]> mean_17_cast_fp16 = reduce_mean(axes = mean_17_axes_0, keep_dims = mean_17_keep_dims_0, x = hidden_states_33_cast_fp16)[name = string("mean_17_cast_fp16")];
            tensor<fp16, [1, 1, 2048]> input_57_cast_fp16 = sub(x = hidden_states_33_cast_fp16, y = mean_17_cast_fp16)[name = string("input_57_cast_fp16")];
            tensor<int32, [1]> var_1012_axes_0 = const()[name = string("op_1012_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2048]> model_model_layers_4_input_layernorm_weight_to_fp16 = const()[name = string("model_model_layers_4_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269200640)))];
            tensor<fp16, [1, 1, 2048]> var_1012_cast_fp16 = layer_norm(axes = var_1012_axes_0, epsilon = var_44_to_fp16, gamma = model_model_layers_4_input_layernorm_weight_to_fp16, x = input_57_cast_fp16)[name = string("op_1012_cast_fp16")];
            tensor<int32, [3]> var_1015 = const()[name = string("op_1015"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1017_axes_0 = const()[name = string("op_1017_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 1]> var_1016 = transpose(perm = var_1015, x = var_1012_cast_fp16)[name = string("transpose_15")];
            tensor<fp16, [1, 2048, 1, 1]> var_1017 = expand_dims(axes = var_1017_axes_0, x = var_1016)[name = string("op_1017")];
            string var_1024_pad_type_0 = const()[name = string("op_1024_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1024_strides_0 = const()[name = string("op_1024_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1024_pad_0 = const()[name = string("op_1024_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1024_dilations_0 = const()[name = string("op_1024_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1024_groups_0 = const()[name = string("op_1024_groups_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 1, 1]> var_1024 = conv(dilations = var_1024_dilations_0, groups = var_1024_groups_0, pad = var_1024_pad_0, pad_type = var_1024_pad_type_0, strides = var_1024_strides_0, weight = model_model_layers_4_self_attn_q_proj_weight_palettized, x = var_1017)[name = string("op_1024")];
            tensor<int32, [4]> var_1025 = const()[name = string("op_1025"), val = tensor<int32, [4]>([1, 32, 1, 64])];
            tensor<fp16, [1, 32, 1, 64]> var_1026 = reshape(shape = var_1025, x = var_1024)[name = string("op_1026")];
            string var_1033_pad_type_0 = const()[name = string("op_1033_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1033_strides_0 = const()[name = string("op_1033_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1033_pad_0 = const()[name = string("op_1033_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1033_dilations_0 = const()[name = string("op_1033_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1033_groups_0 = const()[name = string("op_1033_groups_0"), val = int32(1)];
            tensor<fp16, [1, 512, 1, 1]> var_1033 = conv(dilations = var_1033_dilations_0, groups = var_1033_groups_0, pad = var_1033_pad_0, pad_type = var_1033_pad_type_0, strides = var_1033_strides_0, weight = model_model_layers_4_self_attn_k_proj_weight_palettized, x = var_1017)[name = string("op_1033")];
            tensor<int32, [4]> var_1034 = const()[name = string("op_1034"), val = tensor<int32, [4]>([1, 8, 1, 64])];
            tensor<fp16, [1, 8, 1, 64]> var_1035 = reshape(shape = var_1034, x = var_1033)[name = string("op_1035")];
            string var_1042_pad_type_0 = const()[name = string("op_1042_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1042_strides_0 = const()[name = string("op_1042_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1042_pad_0 = const()[name = string("op_1042_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1042_dilations_0 = const()[name = string("op_1042_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1042_groups_0 = const()[name = string("op_1042_groups_0"), val = int32(1)];
            tensor<fp16, [1, 512, 1, 1]> var_1042 = conv(dilations = var_1042_dilations_0, groups = var_1042_groups_0, pad = var_1042_pad_0, pad_type = var_1042_pad_type_0, strides = var_1042_strides_0, weight = model_model_layers_4_self_attn_v_proj_weight_palettized, x = var_1017)[name = string("op_1042")];
            tensor<int32, [4]> var_1043 = const()[name = string("op_1043"), val = tensor<int32, [4]>([1, 8, 1, 64])];
            tensor<fp16, [1, 8, 1, 64]> var_1044 = reshape(shape = var_1043, x = var_1042)[name = string("op_1044")];
            tensor<int32, [4]> x1_17_begin_0 = const()[name = string("x1_17_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_17_end_0 = const()[name = string("x1_17_end_0"), val = tensor<int32, [4]>([1, 32, 1, 32])];
            tensor<bool, [4]> x1_17_end_mask_0 = const()[name = string("x1_17_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 1, 32]> x1_17 = slice_by_index(begin = x1_17_begin_0, end = x1_17_end_0, end_mask = x1_17_end_mask_0, x = var_1026)[name = string("x1_17")];
            tensor<int32, [4]> x2_17_begin_0 = const()[name = string("x2_17_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_17_end_0 = const()[name = string("x2_17_end_0"), val = tensor<int32, [4]>([1, 32, 1, 64])];
            tensor<bool, [4]> x2_17_end_mask_0 = const()[name = string("x2_17_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 1, 32]> x2_17 = slice_by_index(begin = x2_17_begin_0, end = x2_17_end_0, end_mask = x2_17_end_mask_0, x = var_1026)[name = string("x2_17")];
            tensor<fp16, [1, 32, 1, 32]> var_1058_cast_fp16 = mul(x = x1_17, y = cos_3_cast_fp16)[name = string("op_1058_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_1059_cast_fp16 = mul(x = x2_17, y = sin_3_cast_fp16)[name = string("op_1059_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_1060_cast_fp16 = sub(x = var_1058_cast_fp16, y = var_1059_cast_fp16)[name = string("op_1060_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_1061_cast_fp16 = mul(x = x2_17, y = cos_3_cast_fp16)[name = string("op_1061_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_1062_cast_fp16 = mul(x = x1_17, y = sin_3_cast_fp16)[name = string("op_1062_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_1063_cast_fp16 = add(x = var_1061_cast_fp16, y = var_1062_cast_fp16)[name = string("op_1063_cast_fp16")];
            bool rotated_17_interleave_0 = const()[name = string("rotated_17_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 32, 1, 64]> rotated_17_cast_fp16 = concat(axis = var_49, interleave = rotated_17_interleave_0, values = (var_1060_cast_fp16, var_1063_cast_fp16))[name = string("rotated_17_cast_fp16")];
            tensor<int32, [4]> x1_19_begin_0 = const()[name = string("x1_19_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_19_end_0 = const()[name = string("x1_19_end_0"), val = tensor<int32, [4]>([1, 8, 1, 32])];
            tensor<bool, [4]> x1_19_end_mask_0 = const()[name = string("x1_19_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 1, 32]> x1_19 = slice_by_index(begin = x1_19_begin_0, end = x1_19_end_0, end_mask = x1_19_end_mask_0, x = var_1035)[name = string("x1_19")];
            tensor<int32, [4]> x2_19_begin_0 = const()[name = string("x2_19_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_19_end_0 = const()[name = string("x2_19_end_0"), val = tensor<int32, [4]>([1, 8, 1, 64])];
            tensor<bool, [4]> x2_19_end_mask_0 = const()[name = string("x2_19_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 1, 32]> x2_19 = slice_by_index(begin = x2_19_begin_0, end = x2_19_end_0, end_mask = x2_19_end_mask_0, x = var_1035)[name = string("x2_19")];
            tensor<fp16, [1, 8, 1, 32]> var_1079_cast_fp16 = mul(x = x1_19, y = cos_3_cast_fp16)[name = string("op_1079_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_1080_cast_fp16 = mul(x = x2_19, y = sin_3_cast_fp16)[name = string("op_1080_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_1081_cast_fp16 = sub(x = var_1079_cast_fp16, y = var_1080_cast_fp16)[name = string("op_1081_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_1082_cast_fp16 = mul(x = x2_19, y = cos_3_cast_fp16)[name = string("op_1082_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_1083_cast_fp16 = mul(x = x1_19, y = sin_3_cast_fp16)[name = string("op_1083_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_1084_cast_fp16 = add(x = var_1082_cast_fp16, y = var_1083_cast_fp16)[name = string("op_1084_cast_fp16")];
            bool rotated_19_interleave_0 = const()[name = string("rotated_19_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 8, 1, 64]> rotated_19_cast_fp16 = concat(axis = var_49, interleave = rotated_19_interleave_0, values = (var_1081_cast_fp16, var_1084_cast_fp16))[name = string("rotated_19_cast_fp16")];
            tensor<int32, [1]> expand_dims_48 = const()[name = string("expand_dims_48"), val = tensor<int32, [1]>([4])];
            tensor<int32, [1]> expand_dims_49 = const()[name = string("expand_dims_49"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_51 = const()[name = string("expand_dims_51"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_52 = const()[name = string("expand_dims_52"), val = tensor<int32, [1]>([5])];
            int32 concat_34_axis_0 = const()[name = string("concat_34_axis_0"), val = int32(0)];
            bool concat_34_interleave_0 = const()[name = string("concat_34_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_34 = concat(axis = concat_34_axis_0, interleave = concat_34_interleave_0, values = (expand_dims_48, expand_dims_49, current_pos, expand_dims_51))[name = string("concat_34")];
            tensor<int32, [1]> concat_35_values1_0 = const()[name = string("concat_35_values1_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> concat_35_values3_0 = const()[name = string("concat_35_values3_0"), val = tensor<int32, [1]>([0])];
            int32 concat_35_axis_0 = const()[name = string("concat_35_axis_0"), val = int32(0)];
            bool concat_35_interleave_0 = const()[name = string("concat_35_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_35 = concat(axis = concat_35_axis_0, interleave = concat_35_interleave_0, values = (expand_dims_52, concat_35_values1_0, var_349, concat_35_values3_0))[name = string("concat_35")];
            tensor<int32, [4]> model_model_kv_cache_0_internal_tensor_assign_9_stride_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_9_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_9_begin_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_9_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_9_end_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_9_end_mask_0"), val = tensor<bool, [4]>([false, true, false, true])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_9_squeeze_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_9_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<fp16, [32, 8, 2048, 64]> model_model_kv_cache_0_internal_tensor_assign_9_cast_fp16 = slice_update(begin = concat_34, begin_mask = model_model_kv_cache_0_internal_tensor_assign_9_begin_mask_0, end = concat_35, end_mask = model_model_kv_cache_0_internal_tensor_assign_9_end_mask_0, squeeze_mask = model_model_kv_cache_0_internal_tensor_assign_9_squeeze_mask_0, stride = model_model_kv_cache_0_internal_tensor_assign_9_stride_0, update = rotated_19_cast_fp16, x = coreml_update_state_23)[name = string("model_model_kv_cache_0_internal_tensor_assign_9_cast_fp16")];
            write_state(data = model_model_kv_cache_0_internal_tensor_assign_9_cast_fp16, input = model_model_kv_cache_0)[name = string("coreml_update_state_8_write_state")];
            tensor<fp16, [32, 8, 2048, 64]> coreml_update_state_24 = read_state(input = model_model_kv_cache_0)[name = string("coreml_update_state_8")];
            tensor<int32, [1]> expand_dims_54 = const()[name = string("expand_dims_54"), val = tensor<int32, [1]>([20])];
            tensor<int32, [1]> expand_dims_55 = const()[name = string("expand_dims_55"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_57 = const()[name = string("expand_dims_57"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_58 = const()[name = string("expand_dims_58"), val = tensor<int32, [1]>([21])];
            int32 concat_38_axis_0 = const()[name = string("concat_38_axis_0"), val = int32(0)];
            bool concat_38_interleave_0 = const()[name = string("concat_38_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_38 = concat(axis = concat_38_axis_0, interleave = concat_38_interleave_0, values = (expand_dims_54, expand_dims_55, current_pos, expand_dims_57))[name = string("concat_38")];
            tensor<int32, [1]> concat_39_values1_0 = const()[name = string("concat_39_values1_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> concat_39_values3_0 = const()[name = string("concat_39_values3_0"), val = tensor<int32, [1]>([0])];
            int32 concat_39_axis_0 = const()[name = string("concat_39_axis_0"), val = int32(0)];
            bool concat_39_interleave_0 = const()[name = string("concat_39_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_39 = concat(axis = concat_39_axis_0, interleave = concat_39_interleave_0, values = (expand_dims_58, concat_39_values1_0, var_349, concat_39_values3_0))[name = string("concat_39")];
            tensor<int32, [4]> model_model_kv_cache_0_internal_tensor_assign_10_stride_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_10_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_10_begin_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_10_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_10_end_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_10_end_mask_0"), val = tensor<bool, [4]>([false, true, false, true])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_10_squeeze_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_10_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<fp16, [32, 8, 2048, 64]> model_model_kv_cache_0_internal_tensor_assign_10_cast_fp16 = slice_update(begin = concat_38, begin_mask = model_model_kv_cache_0_internal_tensor_assign_10_begin_mask_0, end = concat_39, end_mask = model_model_kv_cache_0_internal_tensor_assign_10_end_mask_0, squeeze_mask = model_model_kv_cache_0_internal_tensor_assign_10_squeeze_mask_0, stride = model_model_kv_cache_0_internal_tensor_assign_10_stride_0, update = var_1044, x = coreml_update_state_24)[name = string("model_model_kv_cache_0_internal_tensor_assign_10_cast_fp16")];
            write_state(data = model_model_kv_cache_0_internal_tensor_assign_10_cast_fp16, input = model_model_kv_cache_0)[name = string("coreml_update_state_9_write_state")];
            tensor<fp16, [32, 8, 2048, 64]> coreml_update_state_25 = read_state(input = model_model_kv_cache_0)[name = string("coreml_update_state_9")];
            tensor<int32, [4]> var_1104_begin_0 = const()[name = string("op_1104_begin_0"), val = tensor<int32, [4]>([4, 0, 0, 0])];
            tensor<int32, [4]> var_1104_end_0 = const()[name = string("op_1104_end_0"), val = tensor<int32, [4]>([5, 8, 2048, 64])];
            tensor<bool, [4]> var_1104_end_mask_0 = const()[name = string("op_1104_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<fp16, [1, 8, 2048, 64]> var_1104_cast_fp16 = slice_by_index(begin = var_1104_begin_0, end = var_1104_end_0, end_mask = var_1104_end_mask_0, x = coreml_update_state_25)[name = string("op_1104_cast_fp16")];
            tensor<int32, [1]> K_layer_cache_9_axes_0 = const()[name = string("K_layer_cache_9_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<fp16, [8, 2048, 64]> K_layer_cache_9_cast_fp16 = squeeze(axes = K_layer_cache_9_axes_0, x = var_1104_cast_fp16)[name = string("K_layer_cache_9_cast_fp16")];
            tensor<int32, [4]> var_1106_begin_0 = const()[name = string("op_1106_begin_0"), val = tensor<int32, [4]>([20, 0, 0, 0])];
            tensor<int32, [4]> var_1106_end_0 = const()[name = string("op_1106_end_0"), val = tensor<int32, [4]>([21, 8, 2048, 64])];
            tensor<bool, [4]> var_1106_end_mask_0 = const()[name = string("op_1106_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<fp16, [1, 8, 2048, 64]> var_1106_cast_fp16 = slice_by_index(begin = var_1106_begin_0, end = var_1106_end_0, end_mask = var_1106_end_mask_0, x = coreml_update_state_25)[name = string("op_1106_cast_fp16")];
            tensor<int32, [1]> V_layer_cache_9_axes_0 = const()[name = string("V_layer_cache_9_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<fp16, [8, 2048, 64]> V_layer_cache_9_cast_fp16 = squeeze(axes = V_layer_cache_9_axes_0, x = var_1106_cast_fp16)[name = string("V_layer_cache_9_cast_fp16")];
            tensor<int32, [1]> x_123_axes_0 = const()[name = string("x_123_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [8, 1, 2048, 64]> x_123_cast_fp16 = expand_dims(axes = x_123_axes_0, x = K_layer_cache_9_cast_fp16)[name = string("x_123_cast_fp16")];
            tensor<int32, [4]> var_1115 = const()[name = string("op_1115"), val = tensor<int32, [4]>([1, 4, 1, 1])];
            tensor<fp16, [8, 4, 2048, 64]> x_125_cast_fp16 = tile(reps = var_1115, x = x_123_cast_fp16)[name = string("x_125_cast_fp16")];
            tensor<int32, [4]> var_1119 = const()[name = string("op_1119"), val = tensor<int32, [4]>([1, -1, 2048, 64])];
            tensor<fp16, [1, 32, 2048, 64]> key_states_19_cast_fp16 = reshape(shape = var_1119, x = x_125_cast_fp16)[name = string("key_states_19_cast_fp16")];
            tensor<int32, [1]> x_129_axes_0 = const()[name = string("x_129_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [8, 1, 2048, 64]> x_129_cast_fp16 = expand_dims(axes = x_129_axes_0, x = V_layer_cache_9_cast_fp16)[name = string("x_129_cast_fp16")];
            tensor<int32, [4]> var_1122 = const()[name = string("op_1122"), val = tensor<int32, [4]>([1, 4, 1, 1])];
            tensor<fp16, [8, 4, 2048, 64]> x_131_cast_fp16 = tile(reps = var_1122, x = x_129_cast_fp16)[name = string("x_131_cast_fp16")];
            tensor<int32, [4]> var_1126 = const()[name = string("op_1126"), val = tensor<int32, [4]>([1, -1, 2048, 64])];
            tensor<fp16, [1, 32, 2048, 64]> value_states_19_cast_fp16 = reshape(shape = var_1126, x = x_131_cast_fp16)[name = string("value_states_19_cast_fp16")];
            bool var_1129_transpose_x_1 = const()[name = string("op_1129_transpose_x_1"), val = bool(false)];
            bool var_1129_transpose_y_1 = const()[name = string("op_1129_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 2048]> var_1129_cast_fp16 = matmul(transpose_x = var_1129_transpose_x_1, transpose_y = var_1129_transpose_y_1, x = rotated_17_cast_fp16, y = key_states_19_cast_fp16)[name = string("op_1129_cast_fp16")];
            fp16 var_1130_to_fp16 = const()[name = string("op_1130_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 32, 1, 2048]> attn_weights_17_cast_fp16 = mul(x = var_1129_cast_fp16, y = var_1130_to_fp16)[name = string("attn_weights_17_cast_fp16")];
            tensor<fp16, [1, 32, 1, 2048]> x_133_cast_fp16 = add(x = attn_weights_17_cast_fp16, y = causal_mask)[name = string("x_133_cast_fp16")];
            tensor<int32, [1]> reduce_max_4_axes_0 = const()[name = string("reduce_max_4_axes_0"), val = tensor<int32, [1]>([-1])];
            bool reduce_max_4_keep_dims_0 = const()[name = string("reduce_max_4_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1]> reduce_max_4_cast_fp16 = reduce_max(axes = reduce_max_4_axes_0, keep_dims = reduce_max_4_keep_dims_0, x = x_133_cast_fp16)[name = string("reduce_max_4_cast_fp16")];
            tensor<fp16, [1, 32, 1, 2048]> x_135_cast_fp16 = sub(x = x_133_cast_fp16, y = reduce_max_4_cast_fp16)[name = string("x_135_cast_fp16")];
            tensor<fp16, [1, 32, 1, 2048]> exp_x_9_cast_fp16 = exp(x = x_135_cast_fp16)[name = string("exp_x_9_cast_fp16")];
            tensor<int32, [1]> var_1141_axes_0 = const()[name = string("op_1141_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1141_keep_dims_0 = const()[name = string("op_1141_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1]> var_1141_cast_fp16 = reduce_sum(axes = var_1141_axes_0, keep_dims = var_1141_keep_dims_0, x = exp_x_9_cast_fp16)[name = string("op_1141_cast_fp16")];
            tensor<fp16, [1, 32, 1, 2048]> attn_weights_19_cast_fp16 = real_div(x = exp_x_9_cast_fp16, y = var_1141_cast_fp16)[name = string("attn_weights_19_cast_fp16")];
            bool attn_output_25_transpose_x_0 = const()[name = string("attn_output_25_transpose_x_0"), val = bool(false)];
            bool attn_output_25_transpose_y_0 = const()[name = string("attn_output_25_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 32, 1, 64]> attn_output_25_cast_fp16 = matmul(transpose_x = attn_output_25_transpose_x_0, transpose_y = attn_output_25_transpose_y_0, x = attn_weights_19_cast_fp16, y = value_states_19_cast_fp16)[name = string("attn_output_25_cast_fp16")];
            tensor<int32, [4]> var_1144_perm_0 = const()[name = string("op_1144_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1146 = const()[name = string("op_1146"), val = tensor<int32, [3]>([1, 1, 2048])];
            tensor<fp16, [1, 1, 32, 64]> var_1144_cast_fp16 = transpose(perm = var_1144_perm_0, x = attn_output_25_cast_fp16)[name = string("transpose_14")];
            tensor<fp16, [1, 1, 2048]> input_61_cast_fp16 = reshape(shape = var_1146, x = var_1144_cast_fp16)[name = string("input_61_cast_fp16")];
            tensor<fp16, [2048, 2048]> model_model_layers_4_self_attn_o_proj_weight_promoted_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269204800))), lut = tensor<fp16, [256, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271302016))))[name = string("model_model_layers_4_self_attn_o_proj_weight_promoted_to_fp16_palettized")];
            tensor<fp16, [1, 1, 2048]> linear_4_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_4_self_attn_o_proj_weight_promoted_to_fp16_palettized, x = input_61_cast_fp16)[name = string("linear_4_cast_fp16")];
            tensor<fp16, [1, 1, 2048]> hidden_states_37_cast_fp16 = add(x = hidden_states_33_cast_fp16, y = linear_4_cast_fp16)[name = string("hidden_states_37_cast_fp16")];
            tensor<int32, [1]> mean_19_axes_0 = const()[name = string("mean_19_axes_0"), val = tensor<int32, [1]>([-1])];
            bool mean_19_keep_dims_0 = const()[name = string("mean_19_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 1, 1]> mean_19_cast_fp16 = reduce_mean(axes = mean_19_axes_0, keep_dims = mean_19_keep_dims_0, x = hidden_states_37_cast_fp16)[name = string("mean_19_cast_fp16")];
            tensor<fp16, [1, 1, 2048]> input_63_cast_fp16 = sub(x = hidden_states_37_cast_fp16, y = mean_19_cast_fp16)[name = string("input_63_cast_fp16")];
            tensor<int32, [1]> var_1157_axes_0 = const()[name = string("op_1157_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2048]> model_model_layers_4_post_attention_layernorm_weight_to_fp16 = const()[name = string("model_model_layers_4_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271310272)))];
            tensor<fp16, [1, 1, 2048]> var_1157_cast_fp16 = layer_norm(axes = var_1157_axes_0, epsilon = var_44_to_fp16, gamma = model_model_layers_4_post_attention_layernorm_weight_to_fp16, x = input_63_cast_fp16)[name = string("op_1157_cast_fp16")];
            tensor<int32, [3]> var_1164 = const()[name = string("op_1164"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> input_65_axes_0 = const()[name = string("input_65_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 1]> var_1165 = transpose(perm = var_1164, x = var_1157_cast_fp16)[name = string("transpose_13")];
            tensor<fp16, [1, 2048, 1, 1]> input_65 = expand_dims(axes = input_65_axes_0, x = var_1165)[name = string("input_65")];
            string input_67_pad_type_0 = const()[name = string("input_67_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_67_strides_0 = const()[name = string("input_67_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_67_pad_0 = const()[name = string("input_67_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_67_dilations_0 = const()[name = string("input_67_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_67_groups_0 = const()[name = string("input_67_groups_0"), val = int32(1)];
            tensor<fp16, [1, 8192, 1, 1]> input_67 = conv(dilations = input_67_dilations_0, groups = input_67_groups_0, pad = input_67_pad_0, pad_type = input_67_pad_type_0, strides = input_67_strides_0, weight = model_model_layers_4_mlp_gate_proj_weight_palettized, x = input_65)[name = string("input_67")];
            string up_states_9_pad_type_0 = const()[name = string("up_states_9_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> up_states_9_strides_0 = const()[name = string("up_states_9_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> up_states_9_pad_0 = const()[name = string("up_states_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> up_states_9_dilations_0 = const()[name = string("up_states_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 up_states_9_groups_0 = const()[name = string("up_states_9_groups_0"), val = int32(1)];
            tensor<fp16, [1, 8192, 1, 1]> up_states_9 = conv(dilations = up_states_9_dilations_0, groups = up_states_9_groups_0, pad = up_states_9_pad_0, pad_type = up_states_9_pad_type_0, strides = up_states_9_strides_0, weight = model_model_layers_4_mlp_up_proj_weight_palettized, x = input_65)[name = string("up_states_9")];
            tensor<fp16, [1, 8192, 1, 1]> gate_states_9 = silu(x = input_67)[name = string("gate_states_9")];
            tensor<fp16, [1, 8192, 1, 1]> input_69 = mul(x = gate_states_9, y = up_states_9)[name = string("input_69")];
            string hidden_states_39_pad_type_0 = const()[name = string("hidden_states_39_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_39_strides_0 = const()[name = string("hidden_states_39_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_39_pad_0 = const()[name = string("hidden_states_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_39_dilations_0 = const()[name = string("hidden_states_39_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_39_groups_0 = const()[name = string("hidden_states_39_groups_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 1, 1]> hidden_states_39 = conv(dilations = hidden_states_39_dilations_0, groups = hidden_states_39_groups_0, pad = hidden_states_39_pad_0, pad_type = hidden_states_39_pad_type_0, strides = hidden_states_39_strides_0, weight = model_model_layers_4_mlp_down_proj_weight_palettized, x = input_69)[name = string("hidden_states_39")];
            tensor<int32, [1]> var_1187_axes_0 = const()[name = string("op_1187_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 1]> var_1187 = squeeze(axes = var_1187_axes_0, x = hidden_states_39)[name = string("op_1187")];
            tensor<int32, [3]> var_1188 = const()[name = string("op_1188"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 1, 2048]> var_1189 = transpose(perm = var_1188, x = var_1187)[name = string("transpose_12")];
            tensor<fp16, [1, 1, 2048]> hidden_states_41_cast_fp16 = add(x = hidden_states_37_cast_fp16, y = var_1189)[name = string("hidden_states_41_cast_fp16")];
            tensor<int32, [1]> mean_21_axes_0 = const()[name = string("mean_21_axes_0"), val = tensor<int32, [1]>([-1])];
            bool mean_21_keep_dims_0 = const()[name = string("mean_21_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 1, 1]> mean_21_cast_fp16 = reduce_mean(axes = mean_21_axes_0, keep_dims = mean_21_keep_dims_0, x = hidden_states_41_cast_fp16)[name = string("mean_21_cast_fp16")];
            tensor<fp16, [1, 1, 2048]> input_71_cast_fp16 = sub(x = hidden_states_41_cast_fp16, y = mean_21_cast_fp16)[name = string("input_71_cast_fp16")];
            tensor<int32, [1]> var_1197_axes_0 = const()[name = string("op_1197_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2048]> model_model_layers_5_input_layernorm_weight_to_fp16 = const()[name = string("model_model_layers_5_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271314432)))];
            tensor<fp16, [1, 1, 2048]> var_1197_cast_fp16 = layer_norm(axes = var_1197_axes_0, epsilon = var_44_to_fp16, gamma = model_model_layers_5_input_layernorm_weight_to_fp16, x = input_71_cast_fp16)[name = string("op_1197_cast_fp16")];
            tensor<int32, [3]> var_1200 = const()[name = string("op_1200"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1202_axes_0 = const()[name = string("op_1202_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 1]> var_1201 = transpose(perm = var_1200, x = var_1197_cast_fp16)[name = string("transpose_11")];
            tensor<fp16, [1, 2048, 1, 1]> var_1202 = expand_dims(axes = var_1202_axes_0, x = var_1201)[name = string("op_1202")];
            string var_1209_pad_type_0 = const()[name = string("op_1209_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1209_strides_0 = const()[name = string("op_1209_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1209_pad_0 = const()[name = string("op_1209_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1209_dilations_0 = const()[name = string("op_1209_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1209_groups_0 = const()[name = string("op_1209_groups_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 1, 1]> var_1209 = conv(dilations = var_1209_dilations_0, groups = var_1209_groups_0, pad = var_1209_pad_0, pad_type = var_1209_pad_type_0, strides = var_1209_strides_0, weight = model_model_layers_5_self_attn_q_proj_weight_palettized, x = var_1202)[name = string("op_1209")];
            tensor<int32, [4]> var_1210 = const()[name = string("op_1210"), val = tensor<int32, [4]>([1, 32, 1, 64])];
            tensor<fp16, [1, 32, 1, 64]> var_1211 = reshape(shape = var_1210, x = var_1209)[name = string("op_1211")];
            string var_1218_pad_type_0 = const()[name = string("op_1218_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1218_strides_0 = const()[name = string("op_1218_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1218_pad_0 = const()[name = string("op_1218_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1218_dilations_0 = const()[name = string("op_1218_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1218_groups_0 = const()[name = string("op_1218_groups_0"), val = int32(1)];
            tensor<fp16, [1, 512, 1, 1]> var_1218 = conv(dilations = var_1218_dilations_0, groups = var_1218_groups_0, pad = var_1218_pad_0, pad_type = var_1218_pad_type_0, strides = var_1218_strides_0, weight = model_model_layers_5_self_attn_k_proj_weight_palettized, x = var_1202)[name = string("op_1218")];
            tensor<int32, [4]> var_1219 = const()[name = string("op_1219"), val = tensor<int32, [4]>([1, 8, 1, 64])];
            tensor<fp16, [1, 8, 1, 64]> var_1220 = reshape(shape = var_1219, x = var_1218)[name = string("op_1220")];
            string var_1227_pad_type_0 = const()[name = string("op_1227_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1227_strides_0 = const()[name = string("op_1227_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1227_pad_0 = const()[name = string("op_1227_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1227_dilations_0 = const()[name = string("op_1227_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1227_groups_0 = const()[name = string("op_1227_groups_0"), val = int32(1)];
            tensor<fp16, [1, 512, 1, 1]> var_1227 = conv(dilations = var_1227_dilations_0, groups = var_1227_groups_0, pad = var_1227_pad_0, pad_type = var_1227_pad_type_0, strides = var_1227_strides_0, weight = model_model_layers_5_self_attn_v_proj_weight_palettized, x = var_1202)[name = string("op_1227")];
            tensor<int32, [4]> var_1228 = const()[name = string("op_1228"), val = tensor<int32, [4]>([1, 8, 1, 64])];
            tensor<fp16, [1, 8, 1, 64]> var_1229 = reshape(shape = var_1228, x = var_1227)[name = string("op_1229")];
            tensor<int32, [4]> x1_21_begin_0 = const()[name = string("x1_21_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_21_end_0 = const()[name = string("x1_21_end_0"), val = tensor<int32, [4]>([1, 32, 1, 32])];
            tensor<bool, [4]> x1_21_end_mask_0 = const()[name = string("x1_21_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 1, 32]> x1_21 = slice_by_index(begin = x1_21_begin_0, end = x1_21_end_0, end_mask = x1_21_end_mask_0, x = var_1211)[name = string("x1_21")];
            tensor<int32, [4]> x2_21_begin_0 = const()[name = string("x2_21_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_21_end_0 = const()[name = string("x2_21_end_0"), val = tensor<int32, [4]>([1, 32, 1, 64])];
            tensor<bool, [4]> x2_21_end_mask_0 = const()[name = string("x2_21_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 1, 32]> x2_21 = slice_by_index(begin = x2_21_begin_0, end = x2_21_end_0, end_mask = x2_21_end_mask_0, x = var_1211)[name = string("x2_21")];
            tensor<fp16, [1, 32, 1, 32]> var_1243_cast_fp16 = mul(x = x1_21, y = cos_3_cast_fp16)[name = string("op_1243_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_1244_cast_fp16 = mul(x = x2_21, y = sin_3_cast_fp16)[name = string("op_1244_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_1245_cast_fp16 = sub(x = var_1243_cast_fp16, y = var_1244_cast_fp16)[name = string("op_1245_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_1246_cast_fp16 = mul(x = x2_21, y = cos_3_cast_fp16)[name = string("op_1246_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_1247_cast_fp16 = mul(x = x1_21, y = sin_3_cast_fp16)[name = string("op_1247_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_1248_cast_fp16 = add(x = var_1246_cast_fp16, y = var_1247_cast_fp16)[name = string("op_1248_cast_fp16")];
            bool rotated_21_interleave_0 = const()[name = string("rotated_21_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 32, 1, 64]> rotated_21_cast_fp16 = concat(axis = var_49, interleave = rotated_21_interleave_0, values = (var_1245_cast_fp16, var_1248_cast_fp16))[name = string("rotated_21_cast_fp16")];
            tensor<int32, [4]> x1_23_begin_0 = const()[name = string("x1_23_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_23_end_0 = const()[name = string("x1_23_end_0"), val = tensor<int32, [4]>([1, 8, 1, 32])];
            tensor<bool, [4]> x1_23_end_mask_0 = const()[name = string("x1_23_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 1, 32]> x1_23 = slice_by_index(begin = x1_23_begin_0, end = x1_23_end_0, end_mask = x1_23_end_mask_0, x = var_1220)[name = string("x1_23")];
            tensor<int32, [4]> x2_23_begin_0 = const()[name = string("x2_23_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_23_end_0 = const()[name = string("x2_23_end_0"), val = tensor<int32, [4]>([1, 8, 1, 64])];
            tensor<bool, [4]> x2_23_end_mask_0 = const()[name = string("x2_23_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 1, 32]> x2_23 = slice_by_index(begin = x2_23_begin_0, end = x2_23_end_0, end_mask = x2_23_end_mask_0, x = var_1220)[name = string("x2_23")];
            tensor<fp16, [1, 8, 1, 32]> var_1264_cast_fp16 = mul(x = x1_23, y = cos_3_cast_fp16)[name = string("op_1264_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_1265_cast_fp16 = mul(x = x2_23, y = sin_3_cast_fp16)[name = string("op_1265_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_1266_cast_fp16 = sub(x = var_1264_cast_fp16, y = var_1265_cast_fp16)[name = string("op_1266_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_1267_cast_fp16 = mul(x = x2_23, y = cos_3_cast_fp16)[name = string("op_1267_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_1268_cast_fp16 = mul(x = x1_23, y = sin_3_cast_fp16)[name = string("op_1268_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_1269_cast_fp16 = add(x = var_1267_cast_fp16, y = var_1268_cast_fp16)[name = string("op_1269_cast_fp16")];
            bool rotated_23_interleave_0 = const()[name = string("rotated_23_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 8, 1, 64]> rotated_23_cast_fp16 = concat(axis = var_49, interleave = rotated_23_interleave_0, values = (var_1266_cast_fp16, var_1269_cast_fp16))[name = string("rotated_23_cast_fp16")];
            tensor<int32, [1]> expand_dims_60 = const()[name = string("expand_dims_60"), val = tensor<int32, [1]>([5])];
            tensor<int32, [1]> expand_dims_61 = const()[name = string("expand_dims_61"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_63 = const()[name = string("expand_dims_63"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_64 = const()[name = string("expand_dims_64"), val = tensor<int32, [1]>([6])];
            int32 concat_42_axis_0 = const()[name = string("concat_42_axis_0"), val = int32(0)];
            bool concat_42_interleave_0 = const()[name = string("concat_42_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_42 = concat(axis = concat_42_axis_0, interleave = concat_42_interleave_0, values = (expand_dims_60, expand_dims_61, current_pos, expand_dims_63))[name = string("concat_42")];
            tensor<int32, [1]> concat_43_values1_0 = const()[name = string("concat_43_values1_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> concat_43_values3_0 = const()[name = string("concat_43_values3_0"), val = tensor<int32, [1]>([0])];
            int32 concat_43_axis_0 = const()[name = string("concat_43_axis_0"), val = int32(0)];
            bool concat_43_interleave_0 = const()[name = string("concat_43_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_43 = concat(axis = concat_43_axis_0, interleave = concat_43_interleave_0, values = (expand_dims_64, concat_43_values1_0, var_349, concat_43_values3_0))[name = string("concat_43")];
            tensor<int32, [4]> model_model_kv_cache_0_internal_tensor_assign_11_stride_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_11_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_11_begin_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_11_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_11_end_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_11_end_mask_0"), val = tensor<bool, [4]>([false, true, false, true])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_11_squeeze_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_11_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<fp16, [32, 8, 2048, 64]> model_model_kv_cache_0_internal_tensor_assign_11_cast_fp16 = slice_update(begin = concat_42, begin_mask = model_model_kv_cache_0_internal_tensor_assign_11_begin_mask_0, end = concat_43, end_mask = model_model_kv_cache_0_internal_tensor_assign_11_end_mask_0, squeeze_mask = model_model_kv_cache_0_internal_tensor_assign_11_squeeze_mask_0, stride = model_model_kv_cache_0_internal_tensor_assign_11_stride_0, update = rotated_23_cast_fp16, x = coreml_update_state_25)[name = string("model_model_kv_cache_0_internal_tensor_assign_11_cast_fp16")];
            write_state(data = model_model_kv_cache_0_internal_tensor_assign_11_cast_fp16, input = model_model_kv_cache_0)[name = string("coreml_update_state_10_write_state")];
            tensor<fp16, [32, 8, 2048, 64]> coreml_update_state_26 = read_state(input = model_model_kv_cache_0)[name = string("coreml_update_state_10")];
            tensor<int32, [1]> expand_dims_66 = const()[name = string("expand_dims_66"), val = tensor<int32, [1]>([21])];
            tensor<int32, [1]> expand_dims_67 = const()[name = string("expand_dims_67"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_69 = const()[name = string("expand_dims_69"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_70 = const()[name = string("expand_dims_70"), val = tensor<int32, [1]>([22])];
            int32 concat_46_axis_0 = const()[name = string("concat_46_axis_0"), val = int32(0)];
            bool concat_46_interleave_0 = const()[name = string("concat_46_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_46 = concat(axis = concat_46_axis_0, interleave = concat_46_interleave_0, values = (expand_dims_66, expand_dims_67, current_pos, expand_dims_69))[name = string("concat_46")];
            tensor<int32, [1]> concat_47_values1_0 = const()[name = string("concat_47_values1_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> concat_47_values3_0 = const()[name = string("concat_47_values3_0"), val = tensor<int32, [1]>([0])];
            int32 concat_47_axis_0 = const()[name = string("concat_47_axis_0"), val = int32(0)];
            bool concat_47_interleave_0 = const()[name = string("concat_47_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_47 = concat(axis = concat_47_axis_0, interleave = concat_47_interleave_0, values = (expand_dims_70, concat_47_values1_0, var_349, concat_47_values3_0))[name = string("concat_47")];
            tensor<int32, [4]> model_model_kv_cache_0_internal_tensor_assign_12_stride_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_12_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_12_begin_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_12_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_12_end_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_12_end_mask_0"), val = tensor<bool, [4]>([false, true, false, true])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_12_squeeze_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_12_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<fp16, [32, 8, 2048, 64]> model_model_kv_cache_0_internal_tensor_assign_12_cast_fp16 = slice_update(begin = concat_46, begin_mask = model_model_kv_cache_0_internal_tensor_assign_12_begin_mask_0, end = concat_47, end_mask = model_model_kv_cache_0_internal_tensor_assign_12_end_mask_0, squeeze_mask = model_model_kv_cache_0_internal_tensor_assign_12_squeeze_mask_0, stride = model_model_kv_cache_0_internal_tensor_assign_12_stride_0, update = var_1229, x = coreml_update_state_26)[name = string("model_model_kv_cache_0_internal_tensor_assign_12_cast_fp16")];
            write_state(data = model_model_kv_cache_0_internal_tensor_assign_12_cast_fp16, input = model_model_kv_cache_0)[name = string("coreml_update_state_11_write_state")];
            tensor<fp16, [32, 8, 2048, 64]> coreml_update_state_27 = read_state(input = model_model_kv_cache_0)[name = string("coreml_update_state_11")];
            tensor<int32, [4]> var_1289_begin_0 = const()[name = string("op_1289_begin_0"), val = tensor<int32, [4]>([5, 0, 0, 0])];
            tensor<int32, [4]> var_1289_end_0 = const()[name = string("op_1289_end_0"), val = tensor<int32, [4]>([6, 8, 2048, 64])];
            tensor<bool, [4]> var_1289_end_mask_0 = const()[name = string("op_1289_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<fp16, [1, 8, 2048, 64]> var_1289_cast_fp16 = slice_by_index(begin = var_1289_begin_0, end = var_1289_end_0, end_mask = var_1289_end_mask_0, x = coreml_update_state_27)[name = string("op_1289_cast_fp16")];
            tensor<int32, [1]> K_layer_cache_11_axes_0 = const()[name = string("K_layer_cache_11_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<fp16, [8, 2048, 64]> K_layer_cache_11_cast_fp16 = squeeze(axes = K_layer_cache_11_axes_0, x = var_1289_cast_fp16)[name = string("K_layer_cache_11_cast_fp16")];
            tensor<int32, [4]> var_1291_begin_0 = const()[name = string("op_1291_begin_0"), val = tensor<int32, [4]>([21, 0, 0, 0])];
            tensor<int32, [4]> var_1291_end_0 = const()[name = string("op_1291_end_0"), val = tensor<int32, [4]>([22, 8, 2048, 64])];
            tensor<bool, [4]> var_1291_end_mask_0 = const()[name = string("op_1291_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<fp16, [1, 8, 2048, 64]> var_1291_cast_fp16 = slice_by_index(begin = var_1291_begin_0, end = var_1291_end_0, end_mask = var_1291_end_mask_0, x = coreml_update_state_27)[name = string("op_1291_cast_fp16")];
            tensor<int32, [1]> V_layer_cache_11_axes_0 = const()[name = string("V_layer_cache_11_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<fp16, [8, 2048, 64]> V_layer_cache_11_cast_fp16 = squeeze(axes = V_layer_cache_11_axes_0, x = var_1291_cast_fp16)[name = string("V_layer_cache_11_cast_fp16")];
            tensor<int32, [1]> x_151_axes_0 = const()[name = string("x_151_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [8, 1, 2048, 64]> x_151_cast_fp16 = expand_dims(axes = x_151_axes_0, x = K_layer_cache_11_cast_fp16)[name = string("x_151_cast_fp16")];
            tensor<int32, [4]> var_1300 = const()[name = string("op_1300"), val = tensor<int32, [4]>([1, 4, 1, 1])];
            tensor<fp16, [8, 4, 2048, 64]> x_153_cast_fp16 = tile(reps = var_1300, x = x_151_cast_fp16)[name = string("x_153_cast_fp16")];
            tensor<int32, [4]> var_1304 = const()[name = string("op_1304"), val = tensor<int32, [4]>([1, -1, 2048, 64])];
            tensor<fp16, [1, 32, 2048, 64]> key_states_23_cast_fp16 = reshape(shape = var_1304, x = x_153_cast_fp16)[name = string("key_states_23_cast_fp16")];
            tensor<int32, [1]> x_157_axes_0 = const()[name = string("x_157_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [8, 1, 2048, 64]> x_157_cast_fp16 = expand_dims(axes = x_157_axes_0, x = V_layer_cache_11_cast_fp16)[name = string("x_157_cast_fp16")];
            tensor<int32, [4]> var_1307 = const()[name = string("op_1307"), val = tensor<int32, [4]>([1, 4, 1, 1])];
            tensor<fp16, [8, 4, 2048, 64]> x_159_cast_fp16 = tile(reps = var_1307, x = x_157_cast_fp16)[name = string("x_159_cast_fp16")];
            tensor<int32, [4]> var_1311 = const()[name = string("op_1311"), val = tensor<int32, [4]>([1, -1, 2048, 64])];
            tensor<fp16, [1, 32, 2048, 64]> value_states_23_cast_fp16 = reshape(shape = var_1311, x = x_159_cast_fp16)[name = string("value_states_23_cast_fp16")];
            bool var_1314_transpose_x_1 = const()[name = string("op_1314_transpose_x_1"), val = bool(false)];
            bool var_1314_transpose_y_1 = const()[name = string("op_1314_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 2048]> var_1314_cast_fp16 = matmul(transpose_x = var_1314_transpose_x_1, transpose_y = var_1314_transpose_y_1, x = rotated_21_cast_fp16, y = key_states_23_cast_fp16)[name = string("op_1314_cast_fp16")];
            fp16 var_1315_to_fp16 = const()[name = string("op_1315_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 32, 1, 2048]> attn_weights_21_cast_fp16 = mul(x = var_1314_cast_fp16, y = var_1315_to_fp16)[name = string("attn_weights_21_cast_fp16")];
            tensor<fp16, [1, 32, 1, 2048]> x_161_cast_fp16 = add(x = attn_weights_21_cast_fp16, y = causal_mask)[name = string("x_161_cast_fp16")];
            tensor<int32, [1]> reduce_max_5_axes_0 = const()[name = string("reduce_max_5_axes_0"), val = tensor<int32, [1]>([-1])];
            bool reduce_max_5_keep_dims_0 = const()[name = string("reduce_max_5_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1]> reduce_max_5_cast_fp16 = reduce_max(axes = reduce_max_5_axes_0, keep_dims = reduce_max_5_keep_dims_0, x = x_161_cast_fp16)[name = string("reduce_max_5_cast_fp16")];
            tensor<fp16, [1, 32, 1, 2048]> x_163_cast_fp16 = sub(x = x_161_cast_fp16, y = reduce_max_5_cast_fp16)[name = string("x_163_cast_fp16")];
            tensor<fp16, [1, 32, 1, 2048]> exp_x_11_cast_fp16 = exp(x = x_163_cast_fp16)[name = string("exp_x_11_cast_fp16")];
            tensor<int32, [1]> var_1326_axes_0 = const()[name = string("op_1326_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1326_keep_dims_0 = const()[name = string("op_1326_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1]> var_1326_cast_fp16 = reduce_sum(axes = var_1326_axes_0, keep_dims = var_1326_keep_dims_0, x = exp_x_11_cast_fp16)[name = string("op_1326_cast_fp16")];
            tensor<fp16, [1, 32, 1, 2048]> attn_weights_23_cast_fp16 = real_div(x = exp_x_11_cast_fp16, y = var_1326_cast_fp16)[name = string("attn_weights_23_cast_fp16")];
            bool attn_output_31_transpose_x_0 = const()[name = string("attn_output_31_transpose_x_0"), val = bool(false)];
            bool attn_output_31_transpose_y_0 = const()[name = string("attn_output_31_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 32, 1, 64]> attn_output_31_cast_fp16 = matmul(transpose_x = attn_output_31_transpose_x_0, transpose_y = attn_output_31_transpose_y_0, x = attn_weights_23_cast_fp16, y = value_states_23_cast_fp16)[name = string("attn_output_31_cast_fp16")];
            tensor<int32, [4]> var_1329_perm_0 = const()[name = string("op_1329_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1331 = const()[name = string("op_1331"), val = tensor<int32, [3]>([1, 1, 2048])];
            tensor<fp16, [1, 1, 32, 64]> var_1329_cast_fp16 = transpose(perm = var_1329_perm_0, x = attn_output_31_cast_fp16)[name = string("transpose_10")];
            tensor<fp16, [1, 1, 2048]> input_75_cast_fp16 = reshape(shape = var_1331, x = var_1329_cast_fp16)[name = string("input_75_cast_fp16")];
            tensor<fp16, [2048, 2048]> model_model_layers_5_self_attn_o_proj_weight_promoted_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271318592))), lut = tensor<fp16, [256, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(273415808))))[name = string("model_model_layers_5_self_attn_o_proj_weight_promoted_to_fp16_palettized")];
            tensor<fp16, [1, 1, 2048]> linear_5_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_5_self_attn_o_proj_weight_promoted_to_fp16_palettized, x = input_75_cast_fp16)[name = string("linear_5_cast_fp16")];
            tensor<fp16, [1, 1, 2048]> hidden_states_45_cast_fp16 = add(x = hidden_states_41_cast_fp16, y = linear_5_cast_fp16)[name = string("hidden_states_45_cast_fp16")];
            tensor<int32, [1]> mean_23_axes_0 = const()[name = string("mean_23_axes_0"), val = tensor<int32, [1]>([-1])];
            bool mean_23_keep_dims_0 = const()[name = string("mean_23_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 1, 1]> mean_23_cast_fp16 = reduce_mean(axes = mean_23_axes_0, keep_dims = mean_23_keep_dims_0, x = hidden_states_45_cast_fp16)[name = string("mean_23_cast_fp16")];
            tensor<fp16, [1, 1, 2048]> input_77_cast_fp16 = sub(x = hidden_states_45_cast_fp16, y = mean_23_cast_fp16)[name = string("input_77_cast_fp16")];
            tensor<int32, [1]> var_1342_axes_0 = const()[name = string("op_1342_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2048]> model_model_layers_5_post_attention_layernorm_weight_to_fp16 = const()[name = string("model_model_layers_5_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(273424064)))];
            tensor<fp16, [1, 1, 2048]> var_1342_cast_fp16 = layer_norm(axes = var_1342_axes_0, epsilon = var_44_to_fp16, gamma = model_model_layers_5_post_attention_layernorm_weight_to_fp16, x = input_77_cast_fp16)[name = string("op_1342_cast_fp16")];
            tensor<int32, [3]> var_1349 = const()[name = string("op_1349"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> input_79_axes_0 = const()[name = string("input_79_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 1]> var_1350 = transpose(perm = var_1349, x = var_1342_cast_fp16)[name = string("transpose_9")];
            tensor<fp16, [1, 2048, 1, 1]> input_79 = expand_dims(axes = input_79_axes_0, x = var_1350)[name = string("input_79")];
            string input_81_pad_type_0 = const()[name = string("input_81_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_81_strides_0 = const()[name = string("input_81_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_81_pad_0 = const()[name = string("input_81_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_81_dilations_0 = const()[name = string("input_81_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_81_groups_0 = const()[name = string("input_81_groups_0"), val = int32(1)];
            tensor<fp16, [1, 8192, 1, 1]> input_81 = conv(dilations = input_81_dilations_0, groups = input_81_groups_0, pad = input_81_pad_0, pad_type = input_81_pad_type_0, strides = input_81_strides_0, weight = model_model_layers_5_mlp_gate_proj_weight_palettized, x = input_79)[name = string("input_81")];
            string up_states_11_pad_type_0 = const()[name = string("up_states_11_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> up_states_11_strides_0 = const()[name = string("up_states_11_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> up_states_11_pad_0 = const()[name = string("up_states_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> up_states_11_dilations_0 = const()[name = string("up_states_11_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 up_states_11_groups_0 = const()[name = string("up_states_11_groups_0"), val = int32(1)];
            tensor<fp16, [1, 8192, 1, 1]> up_states_11 = conv(dilations = up_states_11_dilations_0, groups = up_states_11_groups_0, pad = up_states_11_pad_0, pad_type = up_states_11_pad_type_0, strides = up_states_11_strides_0, weight = model_model_layers_5_mlp_up_proj_weight_palettized, x = input_79)[name = string("up_states_11")];
            tensor<fp16, [1, 8192, 1, 1]> gate_states_11 = silu(x = input_81)[name = string("gate_states_11")];
            tensor<fp16, [1, 8192, 1, 1]> input_83 = mul(x = gate_states_11, y = up_states_11)[name = string("input_83")];
            string hidden_states_47_pad_type_0 = const()[name = string("hidden_states_47_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_47_strides_0 = const()[name = string("hidden_states_47_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_47_pad_0 = const()[name = string("hidden_states_47_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_47_dilations_0 = const()[name = string("hidden_states_47_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_47_groups_0 = const()[name = string("hidden_states_47_groups_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 1, 1]> hidden_states_47 = conv(dilations = hidden_states_47_dilations_0, groups = hidden_states_47_groups_0, pad = hidden_states_47_pad_0, pad_type = hidden_states_47_pad_type_0, strides = hidden_states_47_strides_0, weight = model_model_layers_5_mlp_down_proj_weight_palettized, x = input_83)[name = string("hidden_states_47")];
            tensor<int32, [1]> var_1372_axes_0 = const()[name = string("op_1372_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 1]> var_1372 = squeeze(axes = var_1372_axes_0, x = hidden_states_47)[name = string("op_1372")];
            tensor<int32, [3]> var_1373 = const()[name = string("op_1373"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 1, 2048]> var_1374 = transpose(perm = var_1373, x = var_1372)[name = string("transpose_8")];
            tensor<fp16, [1, 1, 2048]> hidden_states_49_cast_fp16 = add(x = hidden_states_45_cast_fp16, y = var_1374)[name = string("hidden_states_49_cast_fp16")];
            tensor<int32, [1]> mean_25_axes_0 = const()[name = string("mean_25_axes_0"), val = tensor<int32, [1]>([-1])];
            bool mean_25_keep_dims_0 = const()[name = string("mean_25_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 1, 1]> mean_25_cast_fp16 = reduce_mean(axes = mean_25_axes_0, keep_dims = mean_25_keep_dims_0, x = hidden_states_49_cast_fp16)[name = string("mean_25_cast_fp16")];
            tensor<fp16, [1, 1, 2048]> input_85_cast_fp16 = sub(x = hidden_states_49_cast_fp16, y = mean_25_cast_fp16)[name = string("input_85_cast_fp16")];
            tensor<int32, [1]> var_1382_axes_0 = const()[name = string("op_1382_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2048]> model_model_layers_6_input_layernorm_weight_to_fp16 = const()[name = string("model_model_layers_6_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(273428224)))];
            tensor<fp16, [1, 1, 2048]> var_1382_cast_fp16 = layer_norm(axes = var_1382_axes_0, epsilon = var_44_to_fp16, gamma = model_model_layers_6_input_layernorm_weight_to_fp16, x = input_85_cast_fp16)[name = string("op_1382_cast_fp16")];
            tensor<int32, [3]> var_1385 = const()[name = string("op_1385"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1387_axes_0 = const()[name = string("op_1387_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 1]> var_1386 = transpose(perm = var_1385, x = var_1382_cast_fp16)[name = string("transpose_7")];
            tensor<fp16, [1, 2048, 1, 1]> var_1387 = expand_dims(axes = var_1387_axes_0, x = var_1386)[name = string("op_1387")];
            string var_1394_pad_type_0 = const()[name = string("op_1394_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1394_strides_0 = const()[name = string("op_1394_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1394_pad_0 = const()[name = string("op_1394_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1394_dilations_0 = const()[name = string("op_1394_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1394_groups_0 = const()[name = string("op_1394_groups_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 1, 1]> var_1394 = conv(dilations = var_1394_dilations_0, groups = var_1394_groups_0, pad = var_1394_pad_0, pad_type = var_1394_pad_type_0, strides = var_1394_strides_0, weight = model_model_layers_6_self_attn_q_proj_weight_palettized, x = var_1387)[name = string("op_1394")];
            tensor<int32, [4]> var_1395 = const()[name = string("op_1395"), val = tensor<int32, [4]>([1, 32, 1, 64])];
            tensor<fp16, [1, 32, 1, 64]> var_1396 = reshape(shape = var_1395, x = var_1394)[name = string("op_1396")];
            string var_1403_pad_type_0 = const()[name = string("op_1403_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1403_strides_0 = const()[name = string("op_1403_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1403_pad_0 = const()[name = string("op_1403_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1403_dilations_0 = const()[name = string("op_1403_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1403_groups_0 = const()[name = string("op_1403_groups_0"), val = int32(1)];
            tensor<fp16, [1, 512, 1, 1]> var_1403 = conv(dilations = var_1403_dilations_0, groups = var_1403_groups_0, pad = var_1403_pad_0, pad_type = var_1403_pad_type_0, strides = var_1403_strides_0, weight = model_model_layers_6_self_attn_k_proj_weight_palettized, x = var_1387)[name = string("op_1403")];
            tensor<int32, [4]> var_1404 = const()[name = string("op_1404"), val = tensor<int32, [4]>([1, 8, 1, 64])];
            tensor<fp16, [1, 8, 1, 64]> var_1405 = reshape(shape = var_1404, x = var_1403)[name = string("op_1405")];
            string var_1412_pad_type_0 = const()[name = string("op_1412_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1412_strides_0 = const()[name = string("op_1412_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1412_pad_0 = const()[name = string("op_1412_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1412_dilations_0 = const()[name = string("op_1412_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1412_groups_0 = const()[name = string("op_1412_groups_0"), val = int32(1)];
            tensor<fp16, [1, 512, 1, 1]> var_1412 = conv(dilations = var_1412_dilations_0, groups = var_1412_groups_0, pad = var_1412_pad_0, pad_type = var_1412_pad_type_0, strides = var_1412_strides_0, weight = model_model_layers_6_self_attn_v_proj_weight_palettized, x = var_1387)[name = string("op_1412")];
            tensor<int32, [4]> var_1413 = const()[name = string("op_1413"), val = tensor<int32, [4]>([1, 8, 1, 64])];
            tensor<fp16, [1, 8, 1, 64]> var_1414 = reshape(shape = var_1413, x = var_1412)[name = string("op_1414")];
            tensor<int32, [4]> x1_25_begin_0 = const()[name = string("x1_25_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_25_end_0 = const()[name = string("x1_25_end_0"), val = tensor<int32, [4]>([1, 32, 1, 32])];
            tensor<bool, [4]> x1_25_end_mask_0 = const()[name = string("x1_25_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 1, 32]> x1_25 = slice_by_index(begin = x1_25_begin_0, end = x1_25_end_0, end_mask = x1_25_end_mask_0, x = var_1396)[name = string("x1_25")];
            tensor<int32, [4]> x2_25_begin_0 = const()[name = string("x2_25_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_25_end_0 = const()[name = string("x2_25_end_0"), val = tensor<int32, [4]>([1, 32, 1, 64])];
            tensor<bool, [4]> x2_25_end_mask_0 = const()[name = string("x2_25_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 1, 32]> x2_25 = slice_by_index(begin = x2_25_begin_0, end = x2_25_end_0, end_mask = x2_25_end_mask_0, x = var_1396)[name = string("x2_25")];
            tensor<fp16, [1, 32, 1, 32]> var_1428_cast_fp16 = mul(x = x1_25, y = cos_3_cast_fp16)[name = string("op_1428_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_1429_cast_fp16 = mul(x = x2_25, y = sin_3_cast_fp16)[name = string("op_1429_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_1430_cast_fp16 = sub(x = var_1428_cast_fp16, y = var_1429_cast_fp16)[name = string("op_1430_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_1431_cast_fp16 = mul(x = x2_25, y = cos_3_cast_fp16)[name = string("op_1431_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_1432_cast_fp16 = mul(x = x1_25, y = sin_3_cast_fp16)[name = string("op_1432_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_1433_cast_fp16 = add(x = var_1431_cast_fp16, y = var_1432_cast_fp16)[name = string("op_1433_cast_fp16")];
            bool rotated_25_interleave_0 = const()[name = string("rotated_25_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 32, 1, 64]> rotated_25_cast_fp16 = concat(axis = var_49, interleave = rotated_25_interleave_0, values = (var_1430_cast_fp16, var_1433_cast_fp16))[name = string("rotated_25_cast_fp16")];
            tensor<int32, [4]> x1_27_begin_0 = const()[name = string("x1_27_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_27_end_0 = const()[name = string("x1_27_end_0"), val = tensor<int32, [4]>([1, 8, 1, 32])];
            tensor<bool, [4]> x1_27_end_mask_0 = const()[name = string("x1_27_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 1, 32]> x1_27 = slice_by_index(begin = x1_27_begin_0, end = x1_27_end_0, end_mask = x1_27_end_mask_0, x = var_1405)[name = string("x1_27")];
            tensor<int32, [4]> x2_27_begin_0 = const()[name = string("x2_27_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_27_end_0 = const()[name = string("x2_27_end_0"), val = tensor<int32, [4]>([1, 8, 1, 64])];
            tensor<bool, [4]> x2_27_end_mask_0 = const()[name = string("x2_27_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 1, 32]> x2_27 = slice_by_index(begin = x2_27_begin_0, end = x2_27_end_0, end_mask = x2_27_end_mask_0, x = var_1405)[name = string("x2_27")];
            tensor<fp16, [1, 8, 1, 32]> var_1449_cast_fp16 = mul(x = x1_27, y = cos_3_cast_fp16)[name = string("op_1449_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_1450_cast_fp16 = mul(x = x2_27, y = sin_3_cast_fp16)[name = string("op_1450_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_1451_cast_fp16 = sub(x = var_1449_cast_fp16, y = var_1450_cast_fp16)[name = string("op_1451_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_1452_cast_fp16 = mul(x = x2_27, y = cos_3_cast_fp16)[name = string("op_1452_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_1453_cast_fp16 = mul(x = x1_27, y = sin_3_cast_fp16)[name = string("op_1453_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_1454_cast_fp16 = add(x = var_1452_cast_fp16, y = var_1453_cast_fp16)[name = string("op_1454_cast_fp16")];
            bool rotated_27_interleave_0 = const()[name = string("rotated_27_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 8, 1, 64]> rotated_27_cast_fp16 = concat(axis = var_49, interleave = rotated_27_interleave_0, values = (var_1451_cast_fp16, var_1454_cast_fp16))[name = string("rotated_27_cast_fp16")];
            tensor<int32, [1]> expand_dims_72 = const()[name = string("expand_dims_72"), val = tensor<int32, [1]>([6])];
            tensor<int32, [1]> expand_dims_73 = const()[name = string("expand_dims_73"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_75 = const()[name = string("expand_dims_75"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_76 = const()[name = string("expand_dims_76"), val = tensor<int32, [1]>([7])];
            int32 concat_50_axis_0 = const()[name = string("concat_50_axis_0"), val = int32(0)];
            bool concat_50_interleave_0 = const()[name = string("concat_50_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_50 = concat(axis = concat_50_axis_0, interleave = concat_50_interleave_0, values = (expand_dims_72, expand_dims_73, current_pos, expand_dims_75))[name = string("concat_50")];
            tensor<int32, [1]> concat_51_values1_0 = const()[name = string("concat_51_values1_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> concat_51_values3_0 = const()[name = string("concat_51_values3_0"), val = tensor<int32, [1]>([0])];
            int32 concat_51_axis_0 = const()[name = string("concat_51_axis_0"), val = int32(0)];
            bool concat_51_interleave_0 = const()[name = string("concat_51_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_51 = concat(axis = concat_51_axis_0, interleave = concat_51_interleave_0, values = (expand_dims_76, concat_51_values1_0, var_349, concat_51_values3_0))[name = string("concat_51")];
            tensor<int32, [4]> model_model_kv_cache_0_internal_tensor_assign_13_stride_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_13_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_13_begin_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_13_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_13_end_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_13_end_mask_0"), val = tensor<bool, [4]>([false, true, false, true])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_13_squeeze_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_13_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<fp16, [32, 8, 2048, 64]> model_model_kv_cache_0_internal_tensor_assign_13_cast_fp16 = slice_update(begin = concat_50, begin_mask = model_model_kv_cache_0_internal_tensor_assign_13_begin_mask_0, end = concat_51, end_mask = model_model_kv_cache_0_internal_tensor_assign_13_end_mask_0, squeeze_mask = model_model_kv_cache_0_internal_tensor_assign_13_squeeze_mask_0, stride = model_model_kv_cache_0_internal_tensor_assign_13_stride_0, update = rotated_27_cast_fp16, x = coreml_update_state_27)[name = string("model_model_kv_cache_0_internal_tensor_assign_13_cast_fp16")];
            write_state(data = model_model_kv_cache_0_internal_tensor_assign_13_cast_fp16, input = model_model_kv_cache_0)[name = string("coreml_update_state_12_write_state")];
            tensor<fp16, [32, 8, 2048, 64]> coreml_update_state_28 = read_state(input = model_model_kv_cache_0)[name = string("coreml_update_state_12")];
            tensor<int32, [1]> expand_dims_78 = const()[name = string("expand_dims_78"), val = tensor<int32, [1]>([22])];
            tensor<int32, [1]> expand_dims_79 = const()[name = string("expand_dims_79"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_81 = const()[name = string("expand_dims_81"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_82 = const()[name = string("expand_dims_82"), val = tensor<int32, [1]>([23])];
            int32 concat_54_axis_0 = const()[name = string("concat_54_axis_0"), val = int32(0)];
            bool concat_54_interleave_0 = const()[name = string("concat_54_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_54 = concat(axis = concat_54_axis_0, interleave = concat_54_interleave_0, values = (expand_dims_78, expand_dims_79, current_pos, expand_dims_81))[name = string("concat_54")];
            tensor<int32, [1]> concat_55_values1_0 = const()[name = string("concat_55_values1_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> concat_55_values3_0 = const()[name = string("concat_55_values3_0"), val = tensor<int32, [1]>([0])];
            int32 concat_55_axis_0 = const()[name = string("concat_55_axis_0"), val = int32(0)];
            bool concat_55_interleave_0 = const()[name = string("concat_55_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_55 = concat(axis = concat_55_axis_0, interleave = concat_55_interleave_0, values = (expand_dims_82, concat_55_values1_0, var_349, concat_55_values3_0))[name = string("concat_55")];
            tensor<int32, [4]> model_model_kv_cache_0_internal_tensor_assign_14_stride_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_14_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_14_begin_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_14_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_14_end_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_14_end_mask_0"), val = tensor<bool, [4]>([false, true, false, true])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_14_squeeze_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_14_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<fp16, [32, 8, 2048, 64]> model_model_kv_cache_0_internal_tensor_assign_14_cast_fp16 = slice_update(begin = concat_54, begin_mask = model_model_kv_cache_0_internal_tensor_assign_14_begin_mask_0, end = concat_55, end_mask = model_model_kv_cache_0_internal_tensor_assign_14_end_mask_0, squeeze_mask = model_model_kv_cache_0_internal_tensor_assign_14_squeeze_mask_0, stride = model_model_kv_cache_0_internal_tensor_assign_14_stride_0, update = var_1414, x = coreml_update_state_28)[name = string("model_model_kv_cache_0_internal_tensor_assign_14_cast_fp16")];
            write_state(data = model_model_kv_cache_0_internal_tensor_assign_14_cast_fp16, input = model_model_kv_cache_0)[name = string("coreml_update_state_13_write_state")];
            tensor<fp16, [32, 8, 2048, 64]> coreml_update_state_29 = read_state(input = model_model_kv_cache_0)[name = string("coreml_update_state_13")];
            tensor<int32, [4]> var_1474_begin_0 = const()[name = string("op_1474_begin_0"), val = tensor<int32, [4]>([6, 0, 0, 0])];
            tensor<int32, [4]> var_1474_end_0 = const()[name = string("op_1474_end_0"), val = tensor<int32, [4]>([7, 8, 2048, 64])];
            tensor<bool, [4]> var_1474_end_mask_0 = const()[name = string("op_1474_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<fp16, [1, 8, 2048, 64]> var_1474_cast_fp16 = slice_by_index(begin = var_1474_begin_0, end = var_1474_end_0, end_mask = var_1474_end_mask_0, x = coreml_update_state_29)[name = string("op_1474_cast_fp16")];
            tensor<int32, [1]> K_layer_cache_13_axes_0 = const()[name = string("K_layer_cache_13_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<fp16, [8, 2048, 64]> K_layer_cache_13_cast_fp16 = squeeze(axes = K_layer_cache_13_axes_0, x = var_1474_cast_fp16)[name = string("K_layer_cache_13_cast_fp16")];
            tensor<int32, [4]> var_1476_begin_0 = const()[name = string("op_1476_begin_0"), val = tensor<int32, [4]>([22, 0, 0, 0])];
            tensor<int32, [4]> var_1476_end_0 = const()[name = string("op_1476_end_0"), val = tensor<int32, [4]>([23, 8, 2048, 64])];
            tensor<bool, [4]> var_1476_end_mask_0 = const()[name = string("op_1476_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<fp16, [1, 8, 2048, 64]> var_1476_cast_fp16 = slice_by_index(begin = var_1476_begin_0, end = var_1476_end_0, end_mask = var_1476_end_mask_0, x = coreml_update_state_29)[name = string("op_1476_cast_fp16")];
            tensor<int32, [1]> V_layer_cache_13_axes_0 = const()[name = string("V_layer_cache_13_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<fp16, [8, 2048, 64]> V_layer_cache_13_cast_fp16 = squeeze(axes = V_layer_cache_13_axes_0, x = var_1476_cast_fp16)[name = string("V_layer_cache_13_cast_fp16")];
            tensor<int32, [1]> x_179_axes_0 = const()[name = string("x_179_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [8, 1, 2048, 64]> x_179_cast_fp16 = expand_dims(axes = x_179_axes_0, x = K_layer_cache_13_cast_fp16)[name = string("x_179_cast_fp16")];
            tensor<int32, [4]> var_1485 = const()[name = string("op_1485"), val = tensor<int32, [4]>([1, 4, 1, 1])];
            tensor<fp16, [8, 4, 2048, 64]> x_181_cast_fp16 = tile(reps = var_1485, x = x_179_cast_fp16)[name = string("x_181_cast_fp16")];
            tensor<int32, [4]> var_1489 = const()[name = string("op_1489"), val = tensor<int32, [4]>([1, -1, 2048, 64])];
            tensor<fp16, [1, 32, 2048, 64]> key_states_27_cast_fp16 = reshape(shape = var_1489, x = x_181_cast_fp16)[name = string("key_states_27_cast_fp16")];
            tensor<int32, [1]> x_185_axes_0 = const()[name = string("x_185_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [8, 1, 2048, 64]> x_185_cast_fp16 = expand_dims(axes = x_185_axes_0, x = V_layer_cache_13_cast_fp16)[name = string("x_185_cast_fp16")];
            tensor<int32, [4]> var_1492 = const()[name = string("op_1492"), val = tensor<int32, [4]>([1, 4, 1, 1])];
            tensor<fp16, [8, 4, 2048, 64]> x_187_cast_fp16 = tile(reps = var_1492, x = x_185_cast_fp16)[name = string("x_187_cast_fp16")];
            tensor<int32, [4]> var_1496 = const()[name = string("op_1496"), val = tensor<int32, [4]>([1, -1, 2048, 64])];
            tensor<fp16, [1, 32, 2048, 64]> value_states_27_cast_fp16 = reshape(shape = var_1496, x = x_187_cast_fp16)[name = string("value_states_27_cast_fp16")];
            bool var_1499_transpose_x_1 = const()[name = string("op_1499_transpose_x_1"), val = bool(false)];
            bool var_1499_transpose_y_1 = const()[name = string("op_1499_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 2048]> var_1499_cast_fp16 = matmul(transpose_x = var_1499_transpose_x_1, transpose_y = var_1499_transpose_y_1, x = rotated_25_cast_fp16, y = key_states_27_cast_fp16)[name = string("op_1499_cast_fp16")];
            fp16 var_1500_to_fp16 = const()[name = string("op_1500_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 32, 1, 2048]> attn_weights_25_cast_fp16 = mul(x = var_1499_cast_fp16, y = var_1500_to_fp16)[name = string("attn_weights_25_cast_fp16")];
            tensor<fp16, [1, 32, 1, 2048]> x_189_cast_fp16 = add(x = attn_weights_25_cast_fp16, y = causal_mask)[name = string("x_189_cast_fp16")];
            tensor<int32, [1]> reduce_max_6_axes_0 = const()[name = string("reduce_max_6_axes_0"), val = tensor<int32, [1]>([-1])];
            bool reduce_max_6_keep_dims_0 = const()[name = string("reduce_max_6_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1]> reduce_max_6_cast_fp16 = reduce_max(axes = reduce_max_6_axes_0, keep_dims = reduce_max_6_keep_dims_0, x = x_189_cast_fp16)[name = string("reduce_max_6_cast_fp16")];
            tensor<fp16, [1, 32, 1, 2048]> x_191_cast_fp16 = sub(x = x_189_cast_fp16, y = reduce_max_6_cast_fp16)[name = string("x_191_cast_fp16")];
            tensor<fp16, [1, 32, 1, 2048]> exp_x_13_cast_fp16 = exp(x = x_191_cast_fp16)[name = string("exp_x_13_cast_fp16")];
            tensor<int32, [1]> var_1511_axes_0 = const()[name = string("op_1511_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1511_keep_dims_0 = const()[name = string("op_1511_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1]> var_1511_cast_fp16 = reduce_sum(axes = var_1511_axes_0, keep_dims = var_1511_keep_dims_0, x = exp_x_13_cast_fp16)[name = string("op_1511_cast_fp16")];
            tensor<fp16, [1, 32, 1, 2048]> attn_weights_27_cast_fp16 = real_div(x = exp_x_13_cast_fp16, y = var_1511_cast_fp16)[name = string("attn_weights_27_cast_fp16")];
            bool attn_output_37_transpose_x_0 = const()[name = string("attn_output_37_transpose_x_0"), val = bool(false)];
            bool attn_output_37_transpose_y_0 = const()[name = string("attn_output_37_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 32, 1, 64]> attn_output_37_cast_fp16 = matmul(transpose_x = attn_output_37_transpose_x_0, transpose_y = attn_output_37_transpose_y_0, x = attn_weights_27_cast_fp16, y = value_states_27_cast_fp16)[name = string("attn_output_37_cast_fp16")];
            tensor<int32, [4]> var_1514_perm_0 = const()[name = string("op_1514_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1516 = const()[name = string("op_1516"), val = tensor<int32, [3]>([1, 1, 2048])];
            tensor<fp16, [1, 1, 32, 64]> var_1514_cast_fp16 = transpose(perm = var_1514_perm_0, x = attn_output_37_cast_fp16)[name = string("transpose_6")];
            tensor<fp16, [1, 1, 2048]> input_89_cast_fp16 = reshape(shape = var_1516, x = var_1514_cast_fp16)[name = string("input_89_cast_fp16")];
            tensor<fp16, [2048, 2048]> model_model_layers_6_self_attn_o_proj_weight_promoted_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(273432384))), lut = tensor<fp16, [256, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275529600))))[name = string("model_model_layers_6_self_attn_o_proj_weight_promoted_to_fp16_palettized")];
            tensor<fp16, [1, 1, 2048]> linear_6_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_6_self_attn_o_proj_weight_promoted_to_fp16_palettized, x = input_89_cast_fp16)[name = string("linear_6_cast_fp16")];
            tensor<fp16, [1, 1, 2048]> hidden_states_53_cast_fp16 = add(x = hidden_states_49_cast_fp16, y = linear_6_cast_fp16)[name = string("hidden_states_53_cast_fp16")];
            tensor<int32, [1]> mean_27_axes_0 = const()[name = string("mean_27_axes_0"), val = tensor<int32, [1]>([-1])];
            bool mean_27_keep_dims_0 = const()[name = string("mean_27_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 1, 1]> mean_27_cast_fp16 = reduce_mean(axes = mean_27_axes_0, keep_dims = mean_27_keep_dims_0, x = hidden_states_53_cast_fp16)[name = string("mean_27_cast_fp16")];
            tensor<fp16, [1, 1, 2048]> input_91_cast_fp16 = sub(x = hidden_states_53_cast_fp16, y = mean_27_cast_fp16)[name = string("input_91_cast_fp16")];
            tensor<int32, [1]> var_1527_axes_0 = const()[name = string("op_1527_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2048]> model_model_layers_6_post_attention_layernorm_weight_to_fp16 = const()[name = string("model_model_layers_6_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275537856)))];
            tensor<fp16, [1, 1, 2048]> var_1527_cast_fp16 = layer_norm(axes = var_1527_axes_0, epsilon = var_44_to_fp16, gamma = model_model_layers_6_post_attention_layernorm_weight_to_fp16, x = input_91_cast_fp16)[name = string("op_1527_cast_fp16")];
            tensor<int32, [3]> var_1534 = const()[name = string("op_1534"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> input_93_axes_0 = const()[name = string("input_93_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 1]> var_1535 = transpose(perm = var_1534, x = var_1527_cast_fp16)[name = string("transpose_5")];
            tensor<fp16, [1, 2048, 1, 1]> input_93 = expand_dims(axes = input_93_axes_0, x = var_1535)[name = string("input_93")];
            string input_95_pad_type_0 = const()[name = string("input_95_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_95_strides_0 = const()[name = string("input_95_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_95_pad_0 = const()[name = string("input_95_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_95_dilations_0 = const()[name = string("input_95_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_95_groups_0 = const()[name = string("input_95_groups_0"), val = int32(1)];
            tensor<fp16, [1, 8192, 1, 1]> input_95 = conv(dilations = input_95_dilations_0, groups = input_95_groups_0, pad = input_95_pad_0, pad_type = input_95_pad_type_0, strides = input_95_strides_0, weight = model_model_layers_6_mlp_gate_proj_weight_palettized, x = input_93)[name = string("input_95")];
            string up_states_13_pad_type_0 = const()[name = string("up_states_13_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> up_states_13_strides_0 = const()[name = string("up_states_13_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> up_states_13_pad_0 = const()[name = string("up_states_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> up_states_13_dilations_0 = const()[name = string("up_states_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 up_states_13_groups_0 = const()[name = string("up_states_13_groups_0"), val = int32(1)];
            tensor<fp16, [1, 8192, 1, 1]> up_states_13 = conv(dilations = up_states_13_dilations_0, groups = up_states_13_groups_0, pad = up_states_13_pad_0, pad_type = up_states_13_pad_type_0, strides = up_states_13_strides_0, weight = model_model_layers_6_mlp_up_proj_weight_palettized, x = input_93)[name = string("up_states_13")];
            tensor<fp16, [1, 8192, 1, 1]> gate_states_13 = silu(x = input_95)[name = string("gate_states_13")];
            tensor<fp16, [1, 8192, 1, 1]> input_97 = mul(x = gate_states_13, y = up_states_13)[name = string("input_97")];
            string hidden_states_55_pad_type_0 = const()[name = string("hidden_states_55_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_55_strides_0 = const()[name = string("hidden_states_55_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_55_pad_0 = const()[name = string("hidden_states_55_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_55_dilations_0 = const()[name = string("hidden_states_55_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_55_groups_0 = const()[name = string("hidden_states_55_groups_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 1, 1]> hidden_states_55 = conv(dilations = hidden_states_55_dilations_0, groups = hidden_states_55_groups_0, pad = hidden_states_55_pad_0, pad_type = hidden_states_55_pad_type_0, strides = hidden_states_55_strides_0, weight = model_model_layers_6_mlp_down_proj_weight_palettized, x = input_97)[name = string("hidden_states_55")];
            tensor<int32, [1]> var_1557_axes_0 = const()[name = string("op_1557_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 1]> var_1557 = squeeze(axes = var_1557_axes_0, x = hidden_states_55)[name = string("op_1557")];
            tensor<int32, [3]> var_1558 = const()[name = string("op_1558"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 1, 2048]> var_1559 = transpose(perm = var_1558, x = var_1557)[name = string("transpose_4")];
            tensor<fp16, [1, 1, 2048]> hidden_states_57_cast_fp16 = add(x = hidden_states_53_cast_fp16, y = var_1559)[name = string("hidden_states_57_cast_fp16")];
            tensor<int32, [1]> mean_29_axes_0 = const()[name = string("mean_29_axes_0"), val = tensor<int32, [1]>([-1])];
            bool mean_29_keep_dims_0 = const()[name = string("mean_29_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 1, 1]> mean_29_cast_fp16 = reduce_mean(axes = mean_29_axes_0, keep_dims = mean_29_keep_dims_0, x = hidden_states_57_cast_fp16)[name = string("mean_29_cast_fp16")];
            tensor<fp16, [1, 1, 2048]> input_99_cast_fp16 = sub(x = hidden_states_57_cast_fp16, y = mean_29_cast_fp16)[name = string("input_99_cast_fp16")];
            tensor<int32, [1]> var_1567_axes_0 = const()[name = string("op_1567_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2048]> model_model_layers_7_input_layernorm_weight_to_fp16 = const()[name = string("model_model_layers_7_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275542016)))];
            tensor<fp16, [1, 1, 2048]> var_1567_cast_fp16 = layer_norm(axes = var_1567_axes_0, epsilon = var_44_to_fp16, gamma = model_model_layers_7_input_layernorm_weight_to_fp16, x = input_99_cast_fp16)[name = string("op_1567_cast_fp16")];
            tensor<int32, [3]> var_1570 = const()[name = string("op_1570"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1572_axes_0 = const()[name = string("op_1572_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 1]> var_1571 = transpose(perm = var_1570, x = var_1567_cast_fp16)[name = string("transpose_3")];
            tensor<fp16, [1, 2048, 1, 1]> var_1572 = expand_dims(axes = var_1572_axes_0, x = var_1571)[name = string("op_1572")];
            string var_1579_pad_type_0 = const()[name = string("op_1579_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1579_strides_0 = const()[name = string("op_1579_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1579_pad_0 = const()[name = string("op_1579_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1579_dilations_0 = const()[name = string("op_1579_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1579_groups_0 = const()[name = string("op_1579_groups_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 1, 1]> var_1579 = conv(dilations = var_1579_dilations_0, groups = var_1579_groups_0, pad = var_1579_pad_0, pad_type = var_1579_pad_type_0, strides = var_1579_strides_0, weight = model_model_layers_7_self_attn_q_proj_weight_palettized, x = var_1572)[name = string("op_1579")];
            tensor<int32, [4]> var_1580 = const()[name = string("op_1580"), val = tensor<int32, [4]>([1, 32, 1, 64])];
            tensor<fp16, [1, 32, 1, 64]> var_1581 = reshape(shape = var_1580, x = var_1579)[name = string("op_1581")];
            string var_1588_pad_type_0 = const()[name = string("op_1588_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1588_strides_0 = const()[name = string("op_1588_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1588_pad_0 = const()[name = string("op_1588_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1588_dilations_0 = const()[name = string("op_1588_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1588_groups_0 = const()[name = string("op_1588_groups_0"), val = int32(1)];
            tensor<fp16, [1, 512, 1, 1]> var_1588 = conv(dilations = var_1588_dilations_0, groups = var_1588_groups_0, pad = var_1588_pad_0, pad_type = var_1588_pad_type_0, strides = var_1588_strides_0, weight = model_model_layers_7_self_attn_k_proj_weight_palettized, x = var_1572)[name = string("op_1588")];
            tensor<int32, [4]> var_1589 = const()[name = string("op_1589"), val = tensor<int32, [4]>([1, 8, 1, 64])];
            tensor<fp16, [1, 8, 1, 64]> var_1590 = reshape(shape = var_1589, x = var_1588)[name = string("op_1590")];
            string var_1597_pad_type_0 = const()[name = string("op_1597_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1597_strides_0 = const()[name = string("op_1597_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1597_pad_0 = const()[name = string("op_1597_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1597_dilations_0 = const()[name = string("op_1597_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1597_groups_0 = const()[name = string("op_1597_groups_0"), val = int32(1)];
            tensor<fp16, [1, 512, 1, 1]> var_1597 = conv(dilations = var_1597_dilations_0, groups = var_1597_groups_0, pad = var_1597_pad_0, pad_type = var_1597_pad_type_0, strides = var_1597_strides_0, weight = model_model_layers_7_self_attn_v_proj_weight_palettized, x = var_1572)[name = string("op_1597")];
            tensor<int32, [4]> var_1598 = const()[name = string("op_1598"), val = tensor<int32, [4]>([1, 8, 1, 64])];
            tensor<fp16, [1, 8, 1, 64]> var_1599 = reshape(shape = var_1598, x = var_1597)[name = string("op_1599")];
            tensor<int32, [4]> x1_29_begin_0 = const()[name = string("x1_29_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_29_end_0 = const()[name = string("x1_29_end_0"), val = tensor<int32, [4]>([1, 32, 1, 32])];
            tensor<bool, [4]> x1_29_end_mask_0 = const()[name = string("x1_29_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 1, 32]> x1_29 = slice_by_index(begin = x1_29_begin_0, end = x1_29_end_0, end_mask = x1_29_end_mask_0, x = var_1581)[name = string("x1_29")];
            tensor<int32, [4]> x2_29_begin_0 = const()[name = string("x2_29_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_29_end_0 = const()[name = string("x2_29_end_0"), val = tensor<int32, [4]>([1, 32, 1, 64])];
            tensor<bool, [4]> x2_29_end_mask_0 = const()[name = string("x2_29_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 1, 32]> x2_29 = slice_by_index(begin = x2_29_begin_0, end = x2_29_end_0, end_mask = x2_29_end_mask_0, x = var_1581)[name = string("x2_29")];
            tensor<fp16, [1, 32, 1, 32]> var_1613_cast_fp16 = mul(x = x1_29, y = cos_3_cast_fp16)[name = string("op_1613_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_1614_cast_fp16 = mul(x = x2_29, y = sin_3_cast_fp16)[name = string("op_1614_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_1615_cast_fp16 = sub(x = var_1613_cast_fp16, y = var_1614_cast_fp16)[name = string("op_1615_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_1616_cast_fp16 = mul(x = x2_29, y = cos_3_cast_fp16)[name = string("op_1616_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_1617_cast_fp16 = mul(x = x1_29, y = sin_3_cast_fp16)[name = string("op_1617_cast_fp16")];
            tensor<fp16, [1, 32, 1, 32]> var_1618_cast_fp16 = add(x = var_1616_cast_fp16, y = var_1617_cast_fp16)[name = string("op_1618_cast_fp16")];
            bool rotated_29_interleave_0 = const()[name = string("rotated_29_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 32, 1, 64]> rotated_29_cast_fp16 = concat(axis = var_49, interleave = rotated_29_interleave_0, values = (var_1615_cast_fp16, var_1618_cast_fp16))[name = string("rotated_29_cast_fp16")];
            tensor<int32, [4]> x1_begin_0 = const()[name = string("x1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_end_0 = const()[name = string("x1_end_0"), val = tensor<int32, [4]>([1, 8, 1, 32])];
            tensor<bool, [4]> x1_end_mask_0 = const()[name = string("x1_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 1, 32]> x1 = slice_by_index(begin = x1_begin_0, end = x1_end_0, end_mask = x1_end_mask_0, x = var_1590)[name = string("x1")];
            tensor<int32, [4]> x2_begin_0 = const()[name = string("x2_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_end_0 = const()[name = string("x2_end_0"), val = tensor<int32, [4]>([1, 8, 1, 64])];
            tensor<bool, [4]> x2_end_mask_0 = const()[name = string("x2_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 1, 32]> x2 = slice_by_index(begin = x2_begin_0, end = x2_end_0, end_mask = x2_end_mask_0, x = var_1590)[name = string("x2")];
            tensor<fp16, [1, 8, 1, 32]> var_1634_cast_fp16 = mul(x = x1, y = cos_3_cast_fp16)[name = string("op_1634_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_1635_cast_fp16 = mul(x = x2, y = sin_3_cast_fp16)[name = string("op_1635_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_1636_cast_fp16 = sub(x = var_1634_cast_fp16, y = var_1635_cast_fp16)[name = string("op_1636_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_1637_cast_fp16 = mul(x = x2, y = cos_3_cast_fp16)[name = string("op_1637_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_1638_cast_fp16 = mul(x = x1, y = sin_3_cast_fp16)[name = string("op_1638_cast_fp16")];
            tensor<fp16, [1, 8, 1, 32]> var_1639_cast_fp16 = add(x = var_1637_cast_fp16, y = var_1638_cast_fp16)[name = string("op_1639_cast_fp16")];
            bool rotated_interleave_0 = const()[name = string("rotated_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 8, 1, 64]> rotated_cast_fp16 = concat(axis = var_49, interleave = rotated_interleave_0, values = (var_1636_cast_fp16, var_1639_cast_fp16))[name = string("rotated_cast_fp16")];
            tensor<int32, [1]> expand_dims_84 = const()[name = string("expand_dims_84"), val = tensor<int32, [1]>([7])];
            tensor<int32, [1]> expand_dims_85 = const()[name = string("expand_dims_85"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_87 = const()[name = string("expand_dims_87"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_88 = const()[name = string("expand_dims_88"), val = tensor<int32, [1]>([8])];
            int32 concat_58_axis_0 = const()[name = string("concat_58_axis_0"), val = int32(0)];
            bool concat_58_interleave_0 = const()[name = string("concat_58_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_58 = concat(axis = concat_58_axis_0, interleave = concat_58_interleave_0, values = (expand_dims_84, expand_dims_85, current_pos, expand_dims_87))[name = string("concat_58")];
            tensor<int32, [1]> concat_59_values1_0 = const()[name = string("concat_59_values1_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> concat_59_values3_0 = const()[name = string("concat_59_values3_0"), val = tensor<int32, [1]>([0])];
            int32 concat_59_axis_0 = const()[name = string("concat_59_axis_0"), val = int32(0)];
            bool concat_59_interleave_0 = const()[name = string("concat_59_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_59 = concat(axis = concat_59_axis_0, interleave = concat_59_interleave_0, values = (expand_dims_88, concat_59_values1_0, var_349, concat_59_values3_0))[name = string("concat_59")];
            tensor<int32, [4]> model_model_kv_cache_0_internal_tensor_assign_15_stride_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_15_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_15_begin_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_15_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_15_end_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_15_end_mask_0"), val = tensor<bool, [4]>([false, true, false, true])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_15_squeeze_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_15_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<fp16, [32, 8, 2048, 64]> model_model_kv_cache_0_internal_tensor_assign_15_cast_fp16 = slice_update(begin = concat_58, begin_mask = model_model_kv_cache_0_internal_tensor_assign_15_begin_mask_0, end = concat_59, end_mask = model_model_kv_cache_0_internal_tensor_assign_15_end_mask_0, squeeze_mask = model_model_kv_cache_0_internal_tensor_assign_15_squeeze_mask_0, stride = model_model_kv_cache_0_internal_tensor_assign_15_stride_0, update = rotated_cast_fp16, x = coreml_update_state_29)[name = string("model_model_kv_cache_0_internal_tensor_assign_15_cast_fp16")];
            write_state(data = model_model_kv_cache_0_internal_tensor_assign_15_cast_fp16, input = model_model_kv_cache_0)[name = string("coreml_update_state_14_write_state")];
            tensor<fp16, [32, 8, 2048, 64]> coreml_update_state_30 = read_state(input = model_model_kv_cache_0)[name = string("coreml_update_state_14")];
            tensor<int32, [1]> expand_dims_90 = const()[name = string("expand_dims_90"), val = tensor<int32, [1]>([23])];
            tensor<int32, [1]> expand_dims_91 = const()[name = string("expand_dims_91"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_93 = const()[name = string("expand_dims_93"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_94 = const()[name = string("expand_dims_94"), val = tensor<int32, [1]>([24])];
            int32 concat_62_axis_0 = const()[name = string("concat_62_axis_0"), val = int32(0)];
            bool concat_62_interleave_0 = const()[name = string("concat_62_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_62 = concat(axis = concat_62_axis_0, interleave = concat_62_interleave_0, values = (expand_dims_90, expand_dims_91, current_pos, expand_dims_93))[name = string("concat_62")];
            tensor<int32, [1]> concat_63_values1_0 = const()[name = string("concat_63_values1_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> concat_63_values3_0 = const()[name = string("concat_63_values3_0"), val = tensor<int32, [1]>([0])];
            int32 concat_63_axis_0 = const()[name = string("concat_63_axis_0"), val = int32(0)];
            bool concat_63_interleave_0 = const()[name = string("concat_63_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_63 = concat(axis = concat_63_axis_0, interleave = concat_63_interleave_0, values = (expand_dims_94, concat_63_values1_0, var_349, concat_63_values3_0))[name = string("concat_63")];
            tensor<int32, [4]> model_model_kv_cache_0_internal_tensor_assign_16_stride_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_16_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_16_begin_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_16_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_16_end_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_16_end_mask_0"), val = tensor<bool, [4]>([false, true, false, true])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_16_squeeze_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_16_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<fp16, [32, 8, 2048, 64]> model_model_kv_cache_0_internal_tensor_assign_16_cast_fp16 = slice_update(begin = concat_62, begin_mask = model_model_kv_cache_0_internal_tensor_assign_16_begin_mask_0, end = concat_63, end_mask = model_model_kv_cache_0_internal_tensor_assign_16_end_mask_0, squeeze_mask = model_model_kv_cache_0_internal_tensor_assign_16_squeeze_mask_0, stride = model_model_kv_cache_0_internal_tensor_assign_16_stride_0, update = var_1599, x = coreml_update_state_30)[name = string("model_model_kv_cache_0_internal_tensor_assign_16_cast_fp16")];
            write_state(data = model_model_kv_cache_0_internal_tensor_assign_16_cast_fp16, input = model_model_kv_cache_0)[name = string("coreml_update_state_15_write_state")];
            tensor<fp16, [32, 8, 2048, 64]> coreml_update_state_31 = read_state(input = model_model_kv_cache_0)[name = string("coreml_update_state_15")];
            tensor<int32, [4]> var_1659_begin_0 = const()[name = string("op_1659_begin_0"), val = tensor<int32, [4]>([7, 0, 0, 0])];
            tensor<int32, [4]> var_1659_end_0 = const()[name = string("op_1659_end_0"), val = tensor<int32, [4]>([8, 8, 2048, 64])];
            tensor<bool, [4]> var_1659_end_mask_0 = const()[name = string("op_1659_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<fp16, [1, 8, 2048, 64]> var_1659_cast_fp16 = slice_by_index(begin = var_1659_begin_0, end = var_1659_end_0, end_mask = var_1659_end_mask_0, x = coreml_update_state_31)[name = string("op_1659_cast_fp16")];
            tensor<int32, [1]> K_layer_cache_axes_0 = const()[name = string("K_layer_cache_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<fp16, [8, 2048, 64]> K_layer_cache_cast_fp16 = squeeze(axes = K_layer_cache_axes_0, x = var_1659_cast_fp16)[name = string("K_layer_cache_cast_fp16")];
            tensor<int32, [4]> var_1661_begin_0 = const()[name = string("op_1661_begin_0"), val = tensor<int32, [4]>([23, 0, 0, 0])];
            tensor<int32, [4]> var_1661_end_0 = const()[name = string("op_1661_end_0"), val = tensor<int32, [4]>([24, 8, 2048, 64])];
            tensor<bool, [4]> var_1661_end_mask_0 = const()[name = string("op_1661_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<fp16, [1, 8, 2048, 64]> var_1661_cast_fp16 = slice_by_index(begin = var_1661_begin_0, end = var_1661_end_0, end_mask = var_1661_end_mask_0, x = coreml_update_state_31)[name = string("op_1661_cast_fp16")];
            tensor<int32, [1]> V_layer_cache_axes_0 = const()[name = string("V_layer_cache_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<fp16, [8, 2048, 64]> V_layer_cache_cast_fp16 = squeeze(axes = V_layer_cache_axes_0, x = var_1661_cast_fp16)[name = string("V_layer_cache_cast_fp16")];
            tensor<int32, [1]> x_207_axes_0 = const()[name = string("x_207_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [8, 1, 2048, 64]> x_207_cast_fp16 = expand_dims(axes = x_207_axes_0, x = K_layer_cache_cast_fp16)[name = string("x_207_cast_fp16")];
            tensor<int32, [4]> var_1670 = const()[name = string("op_1670"), val = tensor<int32, [4]>([1, 4, 1, 1])];
            tensor<fp16, [8, 4, 2048, 64]> x_209_cast_fp16 = tile(reps = var_1670, x = x_207_cast_fp16)[name = string("x_209_cast_fp16")];
            tensor<int32, [4]> var_1674 = const()[name = string("op_1674"), val = tensor<int32, [4]>([1, -1, 2048, 64])];
            tensor<fp16, [1, 32, 2048, 64]> key_states_cast_fp16 = reshape(shape = var_1674, x = x_209_cast_fp16)[name = string("key_states_cast_fp16")];
            tensor<int32, [1]> x_213_axes_0 = const()[name = string("x_213_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [8, 1, 2048, 64]> x_213_cast_fp16 = expand_dims(axes = x_213_axes_0, x = V_layer_cache_cast_fp16)[name = string("x_213_cast_fp16")];
            tensor<int32, [4]> var_1677 = const()[name = string("op_1677"), val = tensor<int32, [4]>([1, 4, 1, 1])];
            tensor<fp16, [8, 4, 2048, 64]> x_215_cast_fp16 = tile(reps = var_1677, x = x_213_cast_fp16)[name = string("x_215_cast_fp16")];
            tensor<int32, [4]> var_1681 = const()[name = string("op_1681"), val = tensor<int32, [4]>([1, -1, 2048, 64])];
            tensor<fp16, [1, 32, 2048, 64]> value_states_cast_fp16 = reshape(shape = var_1681, x = x_215_cast_fp16)[name = string("value_states_cast_fp16")];
            bool var_1684_transpose_x_1 = const()[name = string("op_1684_transpose_x_1"), val = bool(false)];
            bool var_1684_transpose_y_1 = const()[name = string("op_1684_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 2048]> var_1684_cast_fp16 = matmul(transpose_x = var_1684_transpose_x_1, transpose_y = var_1684_transpose_y_1, x = rotated_29_cast_fp16, y = key_states_cast_fp16)[name = string("op_1684_cast_fp16")];
            fp16 var_1685_to_fp16 = const()[name = string("op_1685_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 32, 1, 2048]> attn_weights_29_cast_fp16 = mul(x = var_1684_cast_fp16, y = var_1685_to_fp16)[name = string("attn_weights_29_cast_fp16")];
            tensor<fp16, [1, 32, 1, 2048]> x_217_cast_fp16 = add(x = attn_weights_29_cast_fp16, y = causal_mask)[name = string("x_217_cast_fp16")];
            tensor<int32, [1]> reduce_max_7_axes_0 = const()[name = string("reduce_max_7_axes_0"), val = tensor<int32, [1]>([-1])];
            bool reduce_max_7_keep_dims_0 = const()[name = string("reduce_max_7_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1]> reduce_max_7_cast_fp16 = reduce_max(axes = reduce_max_7_axes_0, keep_dims = reduce_max_7_keep_dims_0, x = x_217_cast_fp16)[name = string("reduce_max_7_cast_fp16")];
            tensor<fp16, [1, 32, 1, 2048]> x_219_cast_fp16 = sub(x = x_217_cast_fp16, y = reduce_max_7_cast_fp16)[name = string("x_219_cast_fp16")];
            tensor<fp16, [1, 32, 1, 2048]> exp_x_cast_fp16 = exp(x = x_219_cast_fp16)[name = string("exp_x_cast_fp16")];
            tensor<int32, [1]> var_1696_axes_0 = const()[name = string("op_1696_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1696_keep_dims_0 = const()[name = string("op_1696_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 1, 1]> var_1696_cast_fp16 = reduce_sum(axes = var_1696_axes_0, keep_dims = var_1696_keep_dims_0, x = exp_x_cast_fp16)[name = string("op_1696_cast_fp16")];
            tensor<fp16, [1, 32, 1, 2048]> attn_weights_cast_fp16 = real_div(x = exp_x_cast_fp16, y = var_1696_cast_fp16)[name = string("attn_weights_cast_fp16")];
            bool attn_output_43_transpose_x_0 = const()[name = string("attn_output_43_transpose_x_0"), val = bool(false)];
            bool attn_output_43_transpose_y_0 = const()[name = string("attn_output_43_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 32, 1, 64]> attn_output_43_cast_fp16 = matmul(transpose_x = attn_output_43_transpose_x_0, transpose_y = attn_output_43_transpose_y_0, x = attn_weights_cast_fp16, y = value_states_cast_fp16)[name = string("attn_output_43_cast_fp16")];
            tensor<int32, [4]> var_1699_perm_0 = const()[name = string("op_1699_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1701 = const()[name = string("op_1701"), val = tensor<int32, [3]>([1, 1, 2048])];
            tensor<fp16, [1, 1, 32, 64]> var_1699_cast_fp16 = transpose(perm = var_1699_perm_0, x = attn_output_43_cast_fp16)[name = string("transpose_2")];
            tensor<fp16, [1, 1, 2048]> input_103_cast_fp16 = reshape(shape = var_1701, x = var_1699_cast_fp16)[name = string("input_103_cast_fp16")];
            tensor<fp16, [2048, 2048]> model_model_layers_7_self_attn_o_proj_weight_promoted_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275546176))), lut = tensor<fp16, [256, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277643392))))[name = string("model_model_layers_7_self_attn_o_proj_weight_promoted_to_fp16_palettized")];
            tensor<fp16, [1, 1, 2048]> linear_7_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_7_self_attn_o_proj_weight_promoted_to_fp16_palettized, x = input_103_cast_fp16)[name = string("linear_7_cast_fp16")];
            tensor<fp16, [1, 1, 2048]> hidden_states_61_cast_fp16 = add(x = hidden_states_57_cast_fp16, y = linear_7_cast_fp16)[name = string("hidden_states_61_cast_fp16")];
            tensor<int32, [1]> mean_axes_0 = const()[name = string("mean_axes_0"), val = tensor<int32, [1]>([-1])];
            bool mean_keep_dims_0 = const()[name = string("mean_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 1, 1]> mean_cast_fp16 = reduce_mean(axes = mean_axes_0, keep_dims = mean_keep_dims_0, x = hidden_states_61_cast_fp16)[name = string("mean_cast_fp16")];
            tensor<fp16, [1, 1, 2048]> input_105_cast_fp16 = sub(x = hidden_states_61_cast_fp16, y = mean_cast_fp16)[name = string("input_105_cast_fp16")];
            tensor<int32, [1]> var_1712_axes_0 = const()[name = string("op_1712_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2048]> model_model_layers_7_post_attention_layernorm_weight_to_fp16 = const()[name = string("model_model_layers_7_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277651648)))];
            tensor<fp16, [1, 1, 2048]> var_1712_cast_fp16 = layer_norm(axes = var_1712_axes_0, epsilon = var_44_to_fp16, gamma = model_model_layers_7_post_attention_layernorm_weight_to_fp16, x = input_105_cast_fp16)[name = string("op_1712_cast_fp16")];
            tensor<int32, [3]> var_1719 = const()[name = string("op_1719"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> input_107_axes_0 = const()[name = string("input_107_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 1]> var_1720 = transpose(perm = var_1719, x = var_1712_cast_fp16)[name = string("transpose_1")];
            tensor<fp16, [1, 2048, 1, 1]> input_107 = expand_dims(axes = input_107_axes_0, x = var_1720)[name = string("input_107")];
            string input_109_pad_type_0 = const()[name = string("input_109_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_109_strides_0 = const()[name = string("input_109_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_109_pad_0 = const()[name = string("input_109_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_109_dilations_0 = const()[name = string("input_109_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_109_groups_0 = const()[name = string("input_109_groups_0"), val = int32(1)];
            tensor<fp16, [1, 8192, 1, 1]> input_109 = conv(dilations = input_109_dilations_0, groups = input_109_groups_0, pad = input_109_pad_0, pad_type = input_109_pad_type_0, strides = input_109_strides_0, weight = model_model_layers_7_mlp_gate_proj_weight_palettized, x = input_107)[name = string("input_109")];
            string up_states_pad_type_0 = const()[name = string("up_states_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> up_states_strides_0 = const()[name = string("up_states_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> up_states_pad_0 = const()[name = string("up_states_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> up_states_dilations_0 = const()[name = string("up_states_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 up_states_groups_0 = const()[name = string("up_states_groups_0"), val = int32(1)];
            tensor<fp16, [1, 8192, 1, 1]> up_states = conv(dilations = up_states_dilations_0, groups = up_states_groups_0, pad = up_states_pad_0, pad_type = up_states_pad_type_0, strides = up_states_strides_0, weight = model_model_layers_7_mlp_up_proj_weight_palettized, x = input_107)[name = string("up_states")];
            tensor<fp16, [1, 8192, 1, 1]> gate_states = silu(x = input_109)[name = string("gate_states")];
            tensor<fp16, [1, 8192, 1, 1]> input = mul(x = gate_states, y = up_states)[name = string("input")];
            string hidden_states_pad_type_0 = const()[name = string("hidden_states_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_strides_0 = const()[name = string("hidden_states_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_pad_0 = const()[name = string("hidden_states_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_dilations_0 = const()[name = string("hidden_states_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_groups_0 = const()[name = string("hidden_states_groups_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 1, 1]> hidden_states_1 = conv(dilations = hidden_states_dilations_0, groups = hidden_states_groups_0, pad = hidden_states_pad_0, pad_type = hidden_states_pad_type_0, strides = hidden_states_strides_0, weight = model_model_layers_7_mlp_down_proj_weight_palettized, x = input)[name = string("hidden_states")];
            tensor<int32, [1]> var_1742_axes_0 = const()[name = string("op_1742_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 1]> var_1742 = squeeze(axes = var_1742_axes_0, x = hidden_states_1)[name = string("op_1742")];
            tensor<int32, [3]> var_1743 = const()[name = string("op_1743"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 1, 2048]> var_1744 = transpose(perm = var_1743, x = var_1742)[name = string("transpose_0")];
            tensor<fp16, [1, 1, 2048]> output_hidden_states = add(x = hidden_states_61_cast_fp16, y = var_1744)[name = string("op_1745_cast_fp16")];
            tensor<int32, [1]> position_ids_tmp = identity(x = position_ids)[name = string("position_ids_tmp")];
        } -> (output_hidden_states);
    func prefill<ios18>(tensor<fp16, [1, 1, 64, 2048]> causal_mask, tensor<int32, [1]> current_pos, tensor<fp16, [1, 64, 2048]> hidden_states, state<tensor<fp16, [32, 8, 2048, 64]>> model_model_kv_cache_0, tensor<int32, [64]> position_ids) {
            tensor<fp16, [2048, 2048, 1, 1]> model_model_layers_0_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64))), lut = tensor<fp16, [256, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2097280))))[name = string("model_model_layers_0_self_attn_q_proj_weight_palettized")];
            tensor<fp16, [512, 2048, 1, 1]> model_model_layers_0_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [512, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2105536))), lut = tensor<fp16, [64, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2629888))))[name = string("model_model_layers_0_self_attn_k_proj_weight_palettized")];
            tensor<fp16, [512, 2048, 1, 1]> model_model_layers_0_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [512, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2632000))), lut = tensor<fp16, [64, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3156352))))[name = string("model_model_layers_0_self_attn_v_proj_weight_palettized")];
            tensor<fp16, [8192, 2048, 1, 1]> model_model_layers_0_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [8192, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3158464))), lut = tensor<fp16, [1024, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11547136))))[name = string("model_model_layers_0_mlp_gate_proj_weight_palettized")];
            tensor<fp16, [8192, 2048, 1, 1]> model_model_layers_0_mlp_up_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [8192, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11579968))), lut = tensor<fp16, [1024, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19968640))))[name = string("model_model_layers_0_mlp_up_proj_weight_palettized")];
            tensor<fp16, [2048, 8192, 1, 1]> model_model_layers_0_mlp_down_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 8192, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20001472))), lut = tensor<fp16, [256, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28390144))))[name = string("model_model_layers_0_mlp_down_proj_weight_palettized")];
            tensor<fp16, [2048, 2048, 1, 1]> model_model_layers_1_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28398400))), lut = tensor<fp16, [256, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30495616))))[name = string("model_model_layers_1_self_attn_q_proj_weight_palettized")];
            tensor<fp16, [512, 2048, 1, 1]> model_model_layers_1_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [512, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30503872))), lut = tensor<fp16, [64, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31028224))))[name = string("model_model_layers_1_self_attn_k_proj_weight_palettized")];
            tensor<fp16, [512, 2048, 1, 1]> model_model_layers_1_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [512, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31030336))), lut = tensor<fp16, [64, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31554688))))[name = string("model_model_layers_1_self_attn_v_proj_weight_palettized")];
            tensor<fp16, [8192, 2048, 1, 1]> model_model_layers_1_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [8192, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31556800))), lut = tensor<fp16, [1024, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39945472))))[name = string("model_model_layers_1_mlp_gate_proj_weight_palettized")];
            tensor<fp16, [8192, 2048, 1, 1]> model_model_layers_1_mlp_up_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [8192, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39978304))), lut = tensor<fp16, [1024, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48366976))))[name = string("model_model_layers_1_mlp_up_proj_weight_palettized")];
            tensor<fp16, [2048, 8192, 1, 1]> model_model_layers_1_mlp_down_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 8192, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48399808))), lut = tensor<fp16, [256, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56788480))))[name = string("model_model_layers_1_mlp_down_proj_weight_palettized")];
            tensor<fp16, [2048, 2048, 1, 1]> model_model_layers_2_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56796736))), lut = tensor<fp16, [256, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58893952))))[name = string("model_model_layers_2_self_attn_q_proj_weight_palettized")];
            tensor<fp16, [512, 2048, 1, 1]> model_model_layers_2_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [512, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58902208))), lut = tensor<fp16, [64, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59426560))))[name = string("model_model_layers_2_self_attn_k_proj_weight_palettized")];
            tensor<fp16, [512, 2048, 1, 1]> model_model_layers_2_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [512, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59428672))), lut = tensor<fp16, [64, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59953024))))[name = string("model_model_layers_2_self_attn_v_proj_weight_palettized")];
            tensor<fp16, [8192, 2048, 1, 1]> model_model_layers_2_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [8192, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59955136))), lut = tensor<fp16, [1024, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68343808))))[name = string("model_model_layers_2_mlp_gate_proj_weight_palettized")];
            tensor<fp16, [8192, 2048, 1, 1]> model_model_layers_2_mlp_up_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [8192, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68376640))), lut = tensor<fp16, [1024, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76765312))))[name = string("model_model_layers_2_mlp_up_proj_weight_palettized")];
            tensor<fp16, [2048, 8192, 1, 1]> model_model_layers_2_mlp_down_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 8192, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76798144))), lut = tensor<fp16, [256, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85186816))))[name = string("model_model_layers_2_mlp_down_proj_weight_palettized")];
            tensor<fp16, [2048, 2048, 1, 1]> model_model_layers_3_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85195072))), lut = tensor<fp16, [256, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87292288))))[name = string("model_model_layers_3_self_attn_q_proj_weight_palettized")];
            tensor<fp16, [512, 2048, 1, 1]> model_model_layers_3_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [512, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87300544))), lut = tensor<fp16, [64, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87824896))))[name = string("model_model_layers_3_self_attn_k_proj_weight_palettized")];
            tensor<fp16, [512, 2048, 1, 1]> model_model_layers_3_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [512, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87827008))), lut = tensor<fp16, [64, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88351360))))[name = string("model_model_layers_3_self_attn_v_proj_weight_palettized")];
            tensor<fp16, [8192, 2048, 1, 1]> model_model_layers_3_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [8192, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88353472))), lut = tensor<fp16, [1024, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96742144))))[name = string("model_model_layers_3_mlp_gate_proj_weight_palettized")];
            tensor<fp16, [8192, 2048, 1, 1]> model_model_layers_3_mlp_up_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [8192, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96774976))), lut = tensor<fp16, [1024, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105163648))))[name = string("model_model_layers_3_mlp_up_proj_weight_palettized")];
            tensor<fp16, [2048, 8192, 1, 1]> model_model_layers_3_mlp_down_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 8192, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105196480))), lut = tensor<fp16, [256, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113585152))))[name = string("model_model_layers_3_mlp_down_proj_weight_palettized")];
            tensor<fp16, [2048, 2048, 1, 1]> model_model_layers_4_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113593408))), lut = tensor<fp16, [256, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115690624))))[name = string("model_model_layers_4_self_attn_q_proj_weight_palettized")];
            tensor<fp16, [512, 2048, 1, 1]> model_model_layers_4_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [512, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115698880))), lut = tensor<fp16, [64, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116223232))))[name = string("model_model_layers_4_self_attn_k_proj_weight_palettized")];
            tensor<fp16, [512, 2048, 1, 1]> model_model_layers_4_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [512, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116225344))), lut = tensor<fp16, [64, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116749696))))[name = string("model_model_layers_4_self_attn_v_proj_weight_palettized")];
            tensor<fp16, [8192, 2048, 1, 1]> model_model_layers_4_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [8192, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116751808))), lut = tensor<fp16, [1024, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125140480))))[name = string("model_model_layers_4_mlp_gate_proj_weight_palettized")];
            tensor<fp16, [8192, 2048, 1, 1]> model_model_layers_4_mlp_up_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [8192, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125173312))), lut = tensor<fp16, [1024, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133561984))))[name = string("model_model_layers_4_mlp_up_proj_weight_palettized")];
            tensor<fp16, [2048, 8192, 1, 1]> model_model_layers_4_mlp_down_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 8192, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133594816))), lut = tensor<fp16, [256, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141983488))))[name = string("model_model_layers_4_mlp_down_proj_weight_palettized")];
            tensor<fp16, [2048, 2048, 1, 1]> model_model_layers_5_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141991744))), lut = tensor<fp16, [256, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144088960))))[name = string("model_model_layers_5_self_attn_q_proj_weight_palettized")];
            tensor<fp16, [512, 2048, 1, 1]> model_model_layers_5_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [512, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144097216))), lut = tensor<fp16, [64, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144621568))))[name = string("model_model_layers_5_self_attn_k_proj_weight_palettized")];
            tensor<fp16, [512, 2048, 1, 1]> model_model_layers_5_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [512, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144623680))), lut = tensor<fp16, [64, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145148032))))[name = string("model_model_layers_5_self_attn_v_proj_weight_palettized")];
            tensor<fp16, [8192, 2048, 1, 1]> model_model_layers_5_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [8192, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145150144))), lut = tensor<fp16, [1024, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153538816))))[name = string("model_model_layers_5_mlp_gate_proj_weight_palettized")];
            tensor<fp16, [8192, 2048, 1, 1]> model_model_layers_5_mlp_up_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [8192, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153571648))), lut = tensor<fp16, [1024, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161960320))))[name = string("model_model_layers_5_mlp_up_proj_weight_palettized")];
            tensor<fp16, [2048, 8192, 1, 1]> model_model_layers_5_mlp_down_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 8192, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161993152))), lut = tensor<fp16, [256, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170381824))))[name = string("model_model_layers_5_mlp_down_proj_weight_palettized")];
            tensor<fp16, [2048, 2048, 1, 1]> model_model_layers_6_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170390080))), lut = tensor<fp16, [256, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172487296))))[name = string("model_model_layers_6_self_attn_q_proj_weight_palettized")];
            tensor<fp16, [512, 2048, 1, 1]> model_model_layers_6_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [512, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172495552))), lut = tensor<fp16, [64, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173019904))))[name = string("model_model_layers_6_self_attn_k_proj_weight_palettized")];
            tensor<fp16, [512, 2048, 1, 1]> model_model_layers_6_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [512, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173022016))), lut = tensor<fp16, [64, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173546368))))[name = string("model_model_layers_6_self_attn_v_proj_weight_palettized")];
            tensor<fp16, [8192, 2048, 1, 1]> model_model_layers_6_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [8192, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173548480))), lut = tensor<fp16, [1024, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181937152))))[name = string("model_model_layers_6_mlp_gate_proj_weight_palettized")];
            tensor<fp16, [8192, 2048, 1, 1]> model_model_layers_6_mlp_up_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [8192, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181969984))), lut = tensor<fp16, [1024, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190358656))))[name = string("model_model_layers_6_mlp_up_proj_weight_palettized")];
            tensor<fp16, [2048, 8192, 1, 1]> model_model_layers_6_mlp_down_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 8192, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190391488))), lut = tensor<fp16, [256, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198780160))))[name = string("model_model_layers_6_mlp_down_proj_weight_palettized")];
            tensor<fp16, [2048, 2048, 1, 1]> model_model_layers_7_self_attn_q_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198788416))), lut = tensor<fp16, [256, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200885632))))[name = string("model_model_layers_7_self_attn_q_proj_weight_palettized")];
            tensor<fp16, [512, 2048, 1, 1]> model_model_layers_7_self_attn_k_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [512, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200893888))), lut = tensor<fp16, [64, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201418240))))[name = string("model_model_layers_7_self_attn_k_proj_weight_palettized")];
            tensor<fp16, [512, 2048, 1, 1]> model_model_layers_7_self_attn_v_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [512, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201420352))), lut = tensor<fp16, [64, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201944704))))[name = string("model_model_layers_7_self_attn_v_proj_weight_palettized")];
            tensor<fp16, [8192, 2048, 1, 1]> model_model_layers_7_mlp_gate_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [8192, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201946816))), lut = tensor<fp16, [1024, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210335488))))[name = string("model_model_layers_7_mlp_gate_proj_weight_palettized")];
            tensor<fp16, [8192, 2048, 1, 1]> model_model_layers_7_mlp_up_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [8192, 2048, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210368320))), lut = tensor<fp16, [1024, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(218756992))))[name = string("model_model_layers_7_mlp_up_proj_weight_palettized")];
            tensor<fp16, [2048, 8192, 1, 1]> model_model_layers_7_mlp_down_proj_weight_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 8192, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(218789824))), lut = tensor<fp16, [256, 1, 1, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227178496))))[name = string("model_model_layers_7_mlp_down_proj_weight_palettized")];
            int32 var_44 = const()[name = string("op_44"), val = int32(-1)];
            int32 greater_equal_0_y_0 = const()[name = string("greater_equal_0_y_0"), val = int32(0)];
            tensor<bool, [64]> greater_equal_0 = greater_equal(x = position_ids, y = greater_equal_0_y_0)[name = string("greater_equal_0")];
            int32 slice_by_index_0 = const()[name = string("slice_by_index_0"), val = int32(131072)];
            tensor<int32, [64]> add_0 = add(x = position_ids, y = slice_by_index_0)[name = string("add_0")];
            tensor<int32, [64]> select_0 = select(a = position_ids, b = add_0, cond = greater_equal_0)[name = string("select_0")];
            int32 var_261_axis_0 = const()[name = string("op_261_axis_0"), val = int32(1)];
            int32 var_261_batch_dims_0 = const()[name = string("op_261_batch_dims_0"), val = int32(0)];
            bool var_261_validate_indices_0 = const()[name = string("op_261_validate_indices_0"), val = bool(false)];
            tensor<fp16, [1, 131072, 64]> var_55_to_fp16 = const()[name = string("op_55_to_fp16"), val = tensor<fp16, [1, 131072, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243964032)))];
            tensor<fp16, [1, 64, 64]> var_261_cast_fp16 = gather(axis = var_261_axis_0, batch_dims = var_261_batch_dims_0, indices = select_0, validate_indices = var_261_validate_indices_0, x = var_55_to_fp16)[name = string("op_261_cast_fp16")];
            tensor<int32, [4]> var_262 = const()[name = string("op_262"), val = tensor<int32, [4]>([1, 64, 1, 64])];
            tensor<fp16, [1, 64, 1, 64]> cos_1_cast_fp16 = reshape(shape = var_262, x = var_261_cast_fp16)[name = string("cos_1_cast_fp16")];
            int32 var_266_axis_0 = const()[name = string("op_266_axis_0"), val = int32(1)];
            int32 var_266_batch_dims_0 = const()[name = string("op_266_batch_dims_0"), val = int32(0)];
            bool var_266_validate_indices_0 = const()[name = string("op_266_validate_indices_0"), val = bool(false)];
            tensor<fp16, [1, 131072, 64]> var_50_to_fp16 = const()[name = string("op_50_to_fp16"), val = tensor<fp16, [1, 131072, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227186752)))];
            tensor<fp16, [1, 64, 64]> var_266_cast_fp16 = gather(axis = var_266_axis_0, batch_dims = var_266_batch_dims_0, indices = select_0, validate_indices = var_266_validate_indices_0, x = var_50_to_fp16)[name = string("op_266_cast_fp16")];
            tensor<int32, [4]> var_267 = const()[name = string("op_267"), val = tensor<int32, [4]>([1, 64, 1, 64])];
            tensor<fp16, [1, 64, 1, 64]> sin_1_cast_fp16 = reshape(shape = var_267, x = var_266_cast_fp16)[name = string("sin_1_cast_fp16")];
            tensor<int32, [1]> mean_1_axes_0 = const()[name = string("mean_1_axes_0"), val = tensor<int32, [1]>([-1])];
            bool mean_1_keep_dims_0 = const()[name = string("mean_1_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 64, 1]> mean_1_cast_fp16 = reduce_mean(axes = mean_1_axes_0, keep_dims = mean_1_keep_dims_0, x = hidden_states)[name = string("mean_1_cast_fp16")];
            tensor<fp16, [1, 64, 2048]> input_1_cast_fp16 = sub(x = hidden_states, y = mean_1_cast_fp16)[name = string("input_1_cast_fp16")];
            tensor<int32, [1]> var_277_axes_0 = const()[name = string("op_277_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2048]> model_model_layers_0_input_layernorm_weight_to_fp16 = const()[name = string("model_model_layers_0_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260741312)))];
            fp16 var_46_to_fp16 = const()[name = string("op_46_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 64, 2048]> var_277_cast_fp16 = layer_norm(axes = var_277_axes_0, epsilon = var_46_to_fp16, gamma = model_model_layers_0_input_layernorm_weight_to_fp16, x = input_1_cast_fp16)[name = string("op_277_cast_fp16")];
            tensor<int32, [3]> var_281 = const()[name = string("op_281"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_283_axes_0 = const()[name = string("op_283_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 64]> var_282 = transpose(perm = var_281, x = var_277_cast_fp16)[name = string("transpose_57")];
            tensor<fp16, [1, 2048, 1, 64]> var_283 = expand_dims(axes = var_283_axes_0, x = var_282)[name = string("op_283")];
            string query_states_1_pad_type_0 = const()[name = string("query_states_1_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_states_1_strides_0 = const()[name = string("query_states_1_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_states_1_pad_0 = const()[name = string("query_states_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_states_1_dilations_0 = const()[name = string("query_states_1_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_states_1_groups_0 = const()[name = string("query_states_1_groups_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 1, 64]> query_states_1 = conv(dilations = query_states_1_dilations_0, groups = query_states_1_groups_0, pad = query_states_1_pad_0, pad_type = query_states_1_pad_type_0, strides = query_states_1_strides_0, weight = model_model_layers_0_self_attn_q_proj_weight_palettized, x = var_283)[name = string("query_states_1")];
            string key_states_1_pad_type_0 = const()[name = string("key_states_1_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_states_1_strides_0 = const()[name = string("key_states_1_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_states_1_pad_0 = const()[name = string("key_states_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_states_1_dilations_0 = const()[name = string("key_states_1_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_states_1_groups_0 = const()[name = string("key_states_1_groups_0"), val = int32(1)];
            tensor<fp16, [1, 512, 1, 64]> key_states_1 = conv(dilations = key_states_1_dilations_0, groups = key_states_1_groups_0, pad = key_states_1_pad_0, pad_type = key_states_1_pad_type_0, strides = key_states_1_strides_0, weight = model_model_layers_0_self_attn_k_proj_weight_palettized, x = var_283)[name = string("key_states_1")];
            string value_states_1_pad_type_0 = const()[name = string("value_states_1_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_states_1_strides_0 = const()[name = string("value_states_1_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_states_1_pad_0 = const()[name = string("value_states_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_states_1_dilations_0 = const()[name = string("value_states_1_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_states_1_groups_0 = const()[name = string("value_states_1_groups_0"), val = int32(1)];
            tensor<fp16, [1, 512, 1, 64]> value_states_1 = conv(dilations = value_states_1_dilations_0, groups = value_states_1_groups_0, pad = value_states_1_pad_0, pad_type = value_states_1_pad_type_0, strides = value_states_1_strides_0, weight = model_model_layers_0_self_attn_v_proj_weight_palettized, x = var_283)[name = string("value_states_1")];
            tensor<int32, [4]> var_303 = const()[name = string("op_303"), val = tensor<int32, [4]>([1, 32, 64, 64])];
            tensor<fp16, [1, 32, 64, 64]> var_304 = reshape(shape = var_303, x = query_states_1)[name = string("op_304")];
            tensor<int32, [4]> var_305 = const()[name = string("op_305"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_307 = const()[name = string("op_307"), val = tensor<int32, [4]>([1, 8, 64, 64])];
            tensor<fp16, [1, 8, 64, 64]> var_308 = reshape(shape = var_307, x = key_states_1)[name = string("op_308")];
            tensor<int32, [4]> var_309 = const()[name = string("op_309"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_311 = const()[name = string("op_311"), val = tensor<int32, [4]>([1, 8, 64, 64])];
            tensor<fp16, [1, 8, 64, 64]> var_312 = reshape(shape = var_311, x = value_states_1)[name = string("op_312")];
            tensor<int32, [4]> var_313 = const()[name = string("op_313"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_315 = const()[name = string("op_315"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_317 = const()[name = string("op_317"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> x1_1_begin_0 = const()[name = string("x1_1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_1_end_0 = const()[name = string("x1_1_end_0"), val = tensor<int32, [4]>([1, 32, 64, 32])];
            tensor<bool, [4]> x1_1_end_mask_0 = const()[name = string("x1_1_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 64, 64]> x_1 = transpose(perm = var_305, x = var_304)[name = string("transpose_56")];
            tensor<fp16, [1, 32, 64, 32]> x1_1 = slice_by_index(begin = x1_1_begin_0, end = x1_1_end_0, end_mask = x1_1_end_mask_0, x = x_1)[name = string("x1_1")];
            tensor<int32, [4]> x2_1_begin_0 = const()[name = string("x2_1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_1_end_0 = const()[name = string("x2_1_end_0"), val = tensor<int32, [4]>([1, 32, 64, 64])];
            tensor<bool, [4]> x2_1_end_mask_0 = const()[name = string("x2_1_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 64, 32]> x2_1 = slice_by_index(begin = x2_1_begin_0, end = x2_1_end_0, end_mask = x2_1_end_mask_0, x = x_1)[name = string("x2_1")];
            tensor<int32, [4]> cos_7_begin_0 = const()[name = string("cos_7_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> cos_7_end_0 = const()[name = string("cos_7_end_0"), val = tensor<int32, [4]>([1, 1, 64, 32])];
            tensor<bool, [4]> cos_7_end_mask_0 = const()[name = string("cos_7_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1, 64, 64]> cos_5 = transpose(perm = var_315, x = cos_1_cast_fp16)[name = string("transpose_55")];
            tensor<fp16, [1, 1, 64, 32]> cos_7 = slice_by_index(begin = cos_7_begin_0, end = cos_7_end_0, end_mask = cos_7_end_mask_0, x = cos_5)[name = string("cos_7")];
            tensor<int32, [4]> sin_7_begin_0 = const()[name = string("sin_7_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> sin_7_end_0 = const()[name = string("sin_7_end_0"), val = tensor<int32, [4]>([1, 1, 64, 32])];
            tensor<bool, [4]> sin_7_end_mask_0 = const()[name = string("sin_7_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1, 64, 64]> sin_5 = transpose(perm = var_317, x = sin_1_cast_fp16)[name = string("transpose_54")];
            tensor<fp16, [1, 1, 64, 32]> sin_7 = slice_by_index(begin = sin_7_begin_0, end = sin_7_end_0, end_mask = sin_7_end_mask_0, x = sin_5)[name = string("sin_7")];
            tensor<fp16, [1, 32, 64, 32]> var_331 = mul(x = x1_1, y = cos_7)[name = string("op_331")];
            tensor<fp16, [1, 32, 64, 32]> var_332 = mul(x = x2_1, y = sin_7)[name = string("op_332")];
            tensor<fp16, [1, 32, 64, 32]> var_333 = sub(x = var_331, y = var_332)[name = string("op_333")];
            tensor<fp16, [1, 32, 64, 32]> var_334 = mul(x = x2_1, y = cos_7)[name = string("op_334")];
            tensor<fp16, [1, 32, 64, 32]> var_335 = mul(x = x1_1, y = sin_7)[name = string("op_335")];
            tensor<fp16, [1, 32, 64, 32]> var_336 = add(x = var_334, y = var_335)[name = string("op_336")];
            bool rotated_1_interleave_0 = const()[name = string("rotated_1_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 32, 64, 64]> rotated_1 = concat(axis = var_44, interleave = rotated_1_interleave_0, values = (var_333, var_336))[name = string("rotated_1")];
            tensor<int32, [4]> x1_3_begin_0 = const()[name = string("x1_3_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_3_end_0 = const()[name = string("x1_3_end_0"), val = tensor<int32, [4]>([1, 8, 64, 32])];
            tensor<bool, [4]> x1_3_end_mask_0 = const()[name = string("x1_3_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 64, 64]> x_5 = transpose(perm = var_309, x = var_308)[name = string("transpose_53")];
            tensor<fp16, [1, 8, 64, 32]> x1_3 = slice_by_index(begin = x1_3_begin_0, end = x1_3_end_0, end_mask = x1_3_end_mask_0, x = x_5)[name = string("x1_3")];
            tensor<int32, [4]> x2_3_begin_0 = const()[name = string("x2_3_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_3_end_0 = const()[name = string("x2_3_end_0"), val = tensor<int32, [4]>([1, 8, 64, 64])];
            tensor<bool, [4]> x2_3_end_mask_0 = const()[name = string("x2_3_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 64, 32]> x2_3 = slice_by_index(begin = x2_3_begin_0, end = x2_3_end_0, end_mask = x2_3_end_mask_0, x = x_5)[name = string("x2_3")];
            tensor<fp16, [1, 8, 64, 32]> var_352 = mul(x = x1_3, y = cos_7)[name = string("op_352")];
            tensor<fp16, [1, 8, 64, 32]> var_353 = mul(x = x2_3, y = sin_7)[name = string("op_353")];
            tensor<fp16, [1, 8, 64, 32]> var_354 = sub(x = var_352, y = var_353)[name = string("op_354")];
            tensor<fp16, [1, 8, 64, 32]> var_355 = mul(x = x2_3, y = cos_7)[name = string("op_355")];
            tensor<fp16, [1, 8, 64, 32]> var_356 = mul(x = x1_3, y = sin_7)[name = string("op_356")];
            tensor<fp16, [1, 8, 64, 32]> var_357 = add(x = var_355, y = var_356)[name = string("op_357")];
            bool rotated_3_interleave_0 = const()[name = string("rotated_3_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 8, 64, 64]> rotated_3 = concat(axis = var_44, interleave = rotated_3_interleave_0, values = (var_354, var_357))[name = string("rotated_3")];
            tensor<int32, [1]> seq_length_1 = const()[name = string("seq_length_1"), val = tensor<int32, [1]>([64])];
            tensor<int32, [1]> var_366 = add(x = current_pos, y = seq_length_1)[name = string("op_366")];
            tensor<fp16, [32, 8, 2048, 64]> read_state_0 = read_state(input = model_model_kv_cache_0)[name = string("read_state_0")];
            tensor<int32, [1]> expand_dims_0 = const()[name = string("expand_dims_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_1 = const()[name = string("expand_dims_1"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_3 = const()[name = string("expand_dims_3"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_4 = const()[name = string("expand_dims_4"), val = tensor<int32, [1]>([1])];
            int32 concat_2_axis_0 = const()[name = string("concat_2_axis_0"), val = int32(0)];
            bool concat_2_interleave_0 = const()[name = string("concat_2_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_2 = concat(axis = concat_2_axis_0, interleave = concat_2_interleave_0, values = (expand_dims_0, expand_dims_1, current_pos, expand_dims_3))[name = string("concat_2")];
            tensor<int32, [1]> concat_3_values1_0 = const()[name = string("concat_3_values1_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> concat_3_values3_0 = const()[name = string("concat_3_values3_0"), val = tensor<int32, [1]>([0])];
            int32 concat_3_axis_0 = const()[name = string("concat_3_axis_0"), val = int32(0)];
            bool concat_3_interleave_0 = const()[name = string("concat_3_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_3 = concat(axis = concat_3_axis_0, interleave = concat_3_interleave_0, values = (expand_dims_4, concat_3_values1_0, var_366, concat_3_values3_0))[name = string("concat_3")];
            tensor<int32, [4]> model_model_kv_cache_0_internal_tensor_assign_1_stride_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_1_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_1_begin_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_1_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_1_end_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_1_end_mask_0"), val = tensor<bool, [4]>([false, true, false, true])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_1_squeeze_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_1_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<fp16, [32, 8, 2048, 64]> model_model_kv_cache_0_internal_tensor_assign_1_cast_fp16 = slice_update(begin = concat_2, begin_mask = model_model_kv_cache_0_internal_tensor_assign_1_begin_mask_0, end = concat_3, end_mask = model_model_kv_cache_0_internal_tensor_assign_1_end_mask_0, squeeze_mask = model_model_kv_cache_0_internal_tensor_assign_1_squeeze_mask_0, stride = model_model_kv_cache_0_internal_tensor_assign_1_stride_0, update = rotated_3, x = read_state_0)[name = string("model_model_kv_cache_0_internal_tensor_assign_1_cast_fp16")];
            write_state(data = model_model_kv_cache_0_internal_tensor_assign_1_cast_fp16, input = model_model_kv_cache_0)[name = string("coreml_update_state_16_write_state")];
            tensor<fp16, [32, 8, 2048, 64]> coreml_update_state_16 = read_state(input = model_model_kv_cache_0)[name = string("coreml_update_state_16")];
            tensor<int32, [1]> expand_dims_6 = const()[name = string("expand_dims_6"), val = tensor<int32, [1]>([16])];
            tensor<int32, [1]> expand_dims_7 = const()[name = string("expand_dims_7"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_9 = const()[name = string("expand_dims_9"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_10 = const()[name = string("expand_dims_10"), val = tensor<int32, [1]>([17])];
            int32 concat_6_axis_0 = const()[name = string("concat_6_axis_0"), val = int32(0)];
            bool concat_6_interleave_0 = const()[name = string("concat_6_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_6 = concat(axis = concat_6_axis_0, interleave = concat_6_interleave_0, values = (expand_dims_6, expand_dims_7, current_pos, expand_dims_9))[name = string("concat_6")];
            tensor<int32, [1]> concat_7_values1_0 = const()[name = string("concat_7_values1_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> concat_7_values3_0 = const()[name = string("concat_7_values3_0"), val = tensor<int32, [1]>([0])];
            int32 concat_7_axis_0 = const()[name = string("concat_7_axis_0"), val = int32(0)];
            bool concat_7_interleave_0 = const()[name = string("concat_7_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_7 = concat(axis = concat_7_axis_0, interleave = concat_7_interleave_0, values = (expand_dims_10, concat_7_values1_0, var_366, concat_7_values3_0))[name = string("concat_7")];
            tensor<int32, [4]> model_model_kv_cache_0_internal_tensor_assign_2_stride_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_2_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_2_begin_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_2_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_2_end_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_2_end_mask_0"), val = tensor<bool, [4]>([false, true, false, true])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_2_squeeze_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_2_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<fp16, [1, 8, 64, 64]> value_states_3 = transpose(perm = var_313, x = var_312)[name = string("transpose_52")];
            tensor<fp16, [32, 8, 2048, 64]> model_model_kv_cache_0_internal_tensor_assign_2_cast_fp16 = slice_update(begin = concat_6, begin_mask = model_model_kv_cache_0_internal_tensor_assign_2_begin_mask_0, end = concat_7, end_mask = model_model_kv_cache_0_internal_tensor_assign_2_end_mask_0, squeeze_mask = model_model_kv_cache_0_internal_tensor_assign_2_squeeze_mask_0, stride = model_model_kv_cache_0_internal_tensor_assign_2_stride_0, update = value_states_3, x = coreml_update_state_16)[name = string("model_model_kv_cache_0_internal_tensor_assign_2_cast_fp16")];
            write_state(data = model_model_kv_cache_0_internal_tensor_assign_2_cast_fp16, input = model_model_kv_cache_0)[name = string("coreml_update_state_17_write_state")];
            tensor<fp16, [32, 8, 2048, 64]> coreml_update_state_17 = read_state(input = model_model_kv_cache_0)[name = string("coreml_update_state_17")];
            tensor<int32, [4]> var_380_begin_0 = const()[name = string("op_380_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_380_end_0 = const()[name = string("op_380_end_0"), val = tensor<int32, [4]>([1, 8, 2048, 64])];
            tensor<bool, [4]> var_380_end_mask_0 = const()[name = string("op_380_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<fp16, [1, 8, 2048, 64]> var_380_cast_fp16 = slice_by_index(begin = var_380_begin_0, end = var_380_end_0, end_mask = var_380_end_mask_0, x = coreml_update_state_17)[name = string("op_380_cast_fp16")];
            tensor<int32, [1]> K_layer_cache_1_axes_0 = const()[name = string("K_layer_cache_1_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<fp16, [8, 2048, 64]> K_layer_cache_1_cast_fp16 = squeeze(axes = K_layer_cache_1_axes_0, x = var_380_cast_fp16)[name = string("K_layer_cache_1_cast_fp16")];
            tensor<int32, [4]> var_382_begin_0 = const()[name = string("op_382_begin_0"), val = tensor<int32, [4]>([16, 0, 0, 0])];
            tensor<int32, [4]> var_382_end_0 = const()[name = string("op_382_end_0"), val = tensor<int32, [4]>([17, 8, 2048, 64])];
            tensor<bool, [4]> var_382_end_mask_0 = const()[name = string("op_382_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<fp16, [1, 8, 2048, 64]> var_382_cast_fp16 = slice_by_index(begin = var_382_begin_0, end = var_382_end_0, end_mask = var_382_end_mask_0, x = coreml_update_state_17)[name = string("op_382_cast_fp16")];
            tensor<int32, [1]> V_layer_cache_1_axes_0 = const()[name = string("V_layer_cache_1_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<fp16, [8, 2048, 64]> V_layer_cache_1_cast_fp16 = squeeze(axes = V_layer_cache_1_axes_0, x = var_382_cast_fp16)[name = string("V_layer_cache_1_cast_fp16")];
            tensor<int32, [1]> x_11_axes_0 = const()[name = string("x_11_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [8, 1, 2048, 64]> x_11_cast_fp16 = expand_dims(axes = x_11_axes_0, x = K_layer_cache_1_cast_fp16)[name = string("x_11_cast_fp16")];
            tensor<int32, [4]> var_391 = const()[name = string("op_391"), val = tensor<int32, [4]>([1, 4, 1, 1])];
            tensor<fp16, [8, 4, 2048, 64]> x_13_cast_fp16 = tile(reps = var_391, x = x_11_cast_fp16)[name = string("x_13_cast_fp16")];
            tensor<int32, [4]> var_395 = const()[name = string("op_395"), val = tensor<int32, [4]>([1, -1, 2048, 64])];
            tensor<fp16, [1, 32, 2048, 64]> var_396_cast_fp16 = reshape(shape = var_395, x = x_13_cast_fp16)[name = string("op_396_cast_fp16")];
            tensor<int32, [1]> x_17_axes_0 = const()[name = string("x_17_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [8, 1, 2048, 64]> x_17_cast_fp16 = expand_dims(axes = x_17_axes_0, x = V_layer_cache_1_cast_fp16)[name = string("x_17_cast_fp16")];
            tensor<int32, [4]> var_398 = const()[name = string("op_398"), val = tensor<int32, [4]>([1, 4, 1, 1])];
            tensor<fp16, [8, 4, 2048, 64]> x_19_cast_fp16 = tile(reps = var_398, x = x_17_cast_fp16)[name = string("x_19_cast_fp16")];
            bool var_405_transpose_x_0 = const()[name = string("op_405_transpose_x_0"), val = bool(false)];
            bool var_405_transpose_y_0 = const()[name = string("op_405_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 32, 64, 2048]> var_405_cast_fp16 = matmul(transpose_x = var_405_transpose_x_0, transpose_y = var_405_transpose_y_0, x = rotated_1, y = var_396_cast_fp16)[name = string("op_405_cast_fp16")];
            fp16 var_406_to_fp16 = const()[name = string("op_406_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 32, 64, 2048]> attn_weights_1_cast_fp16 = mul(x = var_405_cast_fp16, y = var_406_to_fp16)[name = string("attn_weights_1_cast_fp16")];
            tensor<fp16, [1, 32, 64, 2048]> x_21_cast_fp16 = add(x = attn_weights_1_cast_fp16, y = causal_mask)[name = string("x_21_cast_fp16")];
            tensor<int32, [1]> reduce_max_0_axes_0 = const()[name = string("reduce_max_0_axes_0"), val = tensor<int32, [1]>([-1])];
            bool reduce_max_0_keep_dims_0 = const()[name = string("reduce_max_0_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 64, 1]> reduce_max_0_cast_fp16 = reduce_max(axes = reduce_max_0_axes_0, keep_dims = reduce_max_0_keep_dims_0, x = x_21_cast_fp16)[name = string("reduce_max_0_cast_fp16")];
            tensor<fp16, [1, 32, 64, 2048]> x_23_cast_fp16 = sub(x = x_21_cast_fp16, y = reduce_max_0_cast_fp16)[name = string("x_23_cast_fp16")];
            tensor<fp16, [1, 32, 64, 2048]> exp_x_1_cast_fp16 = exp(x = x_23_cast_fp16)[name = string("exp_x_1_cast_fp16")];
            tensor<int32, [1]> var_417_axes_0 = const()[name = string("op_417_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_417_keep_dims_0 = const()[name = string("op_417_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 64, 1]> var_417_cast_fp16 = reduce_sum(axes = var_417_axes_0, keep_dims = var_417_keep_dims_0, x = exp_x_1_cast_fp16)[name = string("op_417_cast_fp16")];
            tensor<fp16, [1, 32, 64, 2048]> var_418_cast_fp16 = real_div(x = exp_x_1_cast_fp16, y = var_417_cast_fp16)[name = string("op_418_cast_fp16")];
            tensor<int32, [3]> concat_12 = const()[name = string("concat_12"), val = tensor<int32, [3]>([32, 64, 2048])];
            tensor<fp16, [32, 64, 2048]> reshape_0_cast_fp16 = reshape(shape = concat_12, x = var_418_cast_fp16)[name = string("reshape_0_cast_fp16")];
            tensor<int32, [3]> concat_13 = const()[name = string("concat_13"), val = tensor<int32, [3]>([32, 2048, 64])];
            tensor<fp16, [32, 2048, 64]> reshape_1_cast_fp16 = reshape(shape = concat_13, x = x_19_cast_fp16)[name = string("reshape_1_cast_fp16")];
            bool matmul_0_transpose_x_0 = const()[name = string("matmul_0_transpose_x_0"), val = bool(false)];
            bool matmul_0_transpose_y_0 = const()[name = string("matmul_0_transpose_y_0"), val = bool(false)];
            tensor<fp16, [32, 64, 64]> matmul_0_cast_fp16 = matmul(transpose_x = matmul_0_transpose_x_0, transpose_y = matmul_0_transpose_y_0, x = reshape_0_cast_fp16, y = reshape_1_cast_fp16)[name = string("matmul_0_cast_fp16")];
            tensor<int32, [4]> concat_17 = const()[name = string("concat_17"), val = tensor<int32, [4]>([1, 32, 64, 64])];
            tensor<fp16, [1, 32, 64, 64]> reshape_2_cast_fp16 = reshape(shape = concat_17, x = matmul_0_cast_fp16)[name = string("reshape_2_cast_fp16")];
            tensor<int32, [4]> var_421_perm_0 = const()[name = string("op_421_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_423 = const()[name = string("op_423"), val = tensor<int32, [3]>([1, 64, 2048])];
            tensor<fp16, [1, 64, 32, 64]> var_421_cast_fp16 = transpose(perm = var_421_perm_0, x = reshape_2_cast_fp16)[name = string("transpose_51")];
            tensor<fp16, [1, 64, 2048]> input_5_cast_fp16 = reshape(shape = var_423, x = var_421_cast_fp16)[name = string("input_5_cast_fp16")];
            tensor<fp16, [2048, 2048]> model_model_layers_0_self_attn_o_proj_weight_promoted_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260745472))), lut = tensor<fp16, [256, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262842688))))[name = string("model_model_layers_0_self_attn_o_proj_weight_promoted_to_fp16_palettized")];
            tensor<fp16, [2048]> linear_0_bias_0_to_fp16 = const()[name = string("linear_0_bias_0_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262850944)))];
            tensor<fp16, [1, 64, 2048]> linear_0_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_0_self_attn_o_proj_weight_promoted_to_fp16_palettized, x = input_5_cast_fp16)[name = string("linear_0_cast_fp16")];
            tensor<fp16, [1, 64, 2048]> hidden_states_5_cast_fp16 = add(x = hidden_states, y = linear_0_cast_fp16)[name = string("hidden_states_5_cast_fp16")];
            tensor<int32, [1]> mean_3_axes_0 = const()[name = string("mean_3_axes_0"), val = tensor<int32, [1]>([-1])];
            bool mean_3_keep_dims_0 = const()[name = string("mean_3_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 64, 1]> mean_3_cast_fp16 = reduce_mean(axes = mean_3_axes_0, keep_dims = mean_3_keep_dims_0, x = hidden_states_5_cast_fp16)[name = string("mean_3_cast_fp16")];
            tensor<fp16, [1, 64, 2048]> input_7_cast_fp16 = sub(x = hidden_states_5_cast_fp16, y = mean_3_cast_fp16)[name = string("input_7_cast_fp16")];
            tensor<int32, [1]> var_434_axes_0 = const()[name = string("op_434_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2048]> model_model_layers_0_post_attention_layernorm_weight_to_fp16 = const()[name = string("model_model_layers_0_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262855104)))];
            tensor<fp16, [1, 64, 2048]> var_434_cast_fp16 = layer_norm(axes = var_434_axes_0, epsilon = var_46_to_fp16, gamma = model_model_layers_0_post_attention_layernorm_weight_to_fp16, x = input_7_cast_fp16)[name = string("op_434_cast_fp16")];
            tensor<int32, [3]> var_441 = const()[name = string("op_441"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> input_9_axes_0 = const()[name = string("input_9_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 64]> var_442 = transpose(perm = var_441, x = var_434_cast_fp16)[name = string("transpose_50")];
            tensor<fp16, [1, 2048, 1, 64]> input_9 = expand_dims(axes = input_9_axes_0, x = var_442)[name = string("input_9")];
            string input_11_pad_type_0 = const()[name = string("input_11_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_11_strides_0 = const()[name = string("input_11_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_11_pad_0 = const()[name = string("input_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_11_dilations_0 = const()[name = string("input_11_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_11_groups_0 = const()[name = string("input_11_groups_0"), val = int32(1)];
            tensor<fp16, [1, 8192, 1, 64]> input_11 = conv(dilations = input_11_dilations_0, groups = input_11_groups_0, pad = input_11_pad_0, pad_type = input_11_pad_type_0, strides = input_11_strides_0, weight = model_model_layers_0_mlp_gate_proj_weight_palettized, x = input_9)[name = string("input_11")];
            string up_states_1_pad_type_0 = const()[name = string("up_states_1_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> up_states_1_strides_0 = const()[name = string("up_states_1_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> up_states_1_pad_0 = const()[name = string("up_states_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> up_states_1_dilations_0 = const()[name = string("up_states_1_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 up_states_1_groups_0 = const()[name = string("up_states_1_groups_0"), val = int32(1)];
            tensor<fp16, [1, 8192, 1, 64]> up_states_1 = conv(dilations = up_states_1_dilations_0, groups = up_states_1_groups_0, pad = up_states_1_pad_0, pad_type = up_states_1_pad_type_0, strides = up_states_1_strides_0, weight = model_model_layers_0_mlp_up_proj_weight_palettized, x = input_9)[name = string("up_states_1")];
            tensor<fp16, [1, 8192, 1, 64]> gate_states_1 = silu(x = input_11)[name = string("gate_states_1")];
            tensor<fp16, [1, 8192, 1, 64]> input_13 = mul(x = gate_states_1, y = up_states_1)[name = string("input_13")];
            string hidden_states_7_pad_type_0 = const()[name = string("hidden_states_7_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_7_strides_0 = const()[name = string("hidden_states_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_7_pad_0 = const()[name = string("hidden_states_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_7_dilations_0 = const()[name = string("hidden_states_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_7_groups_0 = const()[name = string("hidden_states_7_groups_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 1, 64]> hidden_states_7 = conv(dilations = hidden_states_7_dilations_0, groups = hidden_states_7_groups_0, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = hidden_states_7_strides_0, weight = model_model_layers_0_mlp_down_proj_weight_palettized, x = input_13)[name = string("hidden_states_7")];
            tensor<int32, [1]> var_464_axes_0 = const()[name = string("op_464_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 64]> var_464 = squeeze(axes = var_464_axes_0, x = hidden_states_7)[name = string("op_464")];
            tensor<int32, [3]> var_465 = const()[name = string("op_465"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 64, 2048]> var_466 = transpose(perm = var_465, x = var_464)[name = string("transpose_49")];
            tensor<fp16, [1, 64, 2048]> hidden_states_9_cast_fp16 = add(x = hidden_states_5_cast_fp16, y = var_466)[name = string("hidden_states_9_cast_fp16")];
            tensor<int32, [1]> mean_5_axes_0 = const()[name = string("mean_5_axes_0"), val = tensor<int32, [1]>([-1])];
            bool mean_5_keep_dims_0 = const()[name = string("mean_5_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 64, 1]> mean_5_cast_fp16 = reduce_mean(axes = mean_5_axes_0, keep_dims = mean_5_keep_dims_0, x = hidden_states_9_cast_fp16)[name = string("mean_5_cast_fp16")];
            tensor<fp16, [1, 64, 2048]> input_15_cast_fp16 = sub(x = hidden_states_9_cast_fp16, y = mean_5_cast_fp16)[name = string("input_15_cast_fp16")];
            tensor<int32, [1]> var_474_axes_0 = const()[name = string("op_474_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2048]> model_model_layers_1_input_layernorm_weight_to_fp16 = const()[name = string("model_model_layers_1_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262859264)))];
            tensor<fp16, [1, 64, 2048]> var_474_cast_fp16 = layer_norm(axes = var_474_axes_0, epsilon = var_46_to_fp16, gamma = model_model_layers_1_input_layernorm_weight_to_fp16, x = input_15_cast_fp16)[name = string("op_474_cast_fp16")];
            tensor<int32, [3]> var_478 = const()[name = string("op_478"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_480_axes_0 = const()[name = string("op_480_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 64]> var_479 = transpose(perm = var_478, x = var_474_cast_fp16)[name = string("transpose_48")];
            tensor<fp16, [1, 2048, 1, 64]> var_480 = expand_dims(axes = var_480_axes_0, x = var_479)[name = string("op_480")];
            string query_states_5_pad_type_0 = const()[name = string("query_states_5_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_states_5_strides_0 = const()[name = string("query_states_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_states_5_pad_0 = const()[name = string("query_states_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_states_5_dilations_0 = const()[name = string("query_states_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_states_5_groups_0 = const()[name = string("query_states_5_groups_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 1, 64]> query_states_5 = conv(dilations = query_states_5_dilations_0, groups = query_states_5_groups_0, pad = query_states_5_pad_0, pad_type = query_states_5_pad_type_0, strides = query_states_5_strides_0, weight = model_model_layers_1_self_attn_q_proj_weight_palettized, x = var_480)[name = string("query_states_5")];
            string key_states_7_pad_type_0 = const()[name = string("key_states_7_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_states_7_strides_0 = const()[name = string("key_states_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_states_7_pad_0 = const()[name = string("key_states_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_states_7_dilations_0 = const()[name = string("key_states_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_states_7_groups_0 = const()[name = string("key_states_7_groups_0"), val = int32(1)];
            tensor<fp16, [1, 512, 1, 64]> key_states_7 = conv(dilations = key_states_7_dilations_0, groups = key_states_7_groups_0, pad = key_states_7_pad_0, pad_type = key_states_7_pad_type_0, strides = key_states_7_strides_0, weight = model_model_layers_1_self_attn_k_proj_weight_palettized, x = var_480)[name = string("key_states_7")];
            string value_states_7_pad_type_0 = const()[name = string("value_states_7_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_states_7_strides_0 = const()[name = string("value_states_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_states_7_pad_0 = const()[name = string("value_states_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_states_7_dilations_0 = const()[name = string("value_states_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_states_7_groups_0 = const()[name = string("value_states_7_groups_0"), val = int32(1)];
            tensor<fp16, [1, 512, 1, 64]> value_states_7 = conv(dilations = value_states_7_dilations_0, groups = value_states_7_groups_0, pad = value_states_7_pad_0, pad_type = value_states_7_pad_type_0, strides = value_states_7_strides_0, weight = model_model_layers_1_self_attn_v_proj_weight_palettized, x = var_480)[name = string("value_states_7")];
            tensor<int32, [4]> var_500 = const()[name = string("op_500"), val = tensor<int32, [4]>([1, 32, 64, 64])];
            tensor<fp16, [1, 32, 64, 64]> var_501 = reshape(shape = var_500, x = query_states_5)[name = string("op_501")];
            tensor<int32, [4]> var_502 = const()[name = string("op_502"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_504 = const()[name = string("op_504"), val = tensor<int32, [4]>([1, 8, 64, 64])];
            tensor<fp16, [1, 8, 64, 64]> var_505 = reshape(shape = var_504, x = key_states_7)[name = string("op_505")];
            tensor<int32, [4]> var_506 = const()[name = string("op_506"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_508 = const()[name = string("op_508"), val = tensor<int32, [4]>([1, 8, 64, 64])];
            tensor<fp16, [1, 8, 64, 64]> var_509 = reshape(shape = var_508, x = value_states_7)[name = string("op_509")];
            tensor<int32, [4]> var_510 = const()[name = string("op_510"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> x1_5_begin_0 = const()[name = string("x1_5_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_5_end_0 = const()[name = string("x1_5_end_0"), val = tensor<int32, [4]>([1, 32, 64, 32])];
            tensor<bool, [4]> x1_5_end_mask_0 = const()[name = string("x1_5_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 64, 64]> x_29 = transpose(perm = var_502, x = var_501)[name = string("transpose_47")];
            tensor<fp16, [1, 32, 64, 32]> x1_5 = slice_by_index(begin = x1_5_begin_0, end = x1_5_end_0, end_mask = x1_5_end_mask_0, x = x_29)[name = string("x1_5")];
            tensor<int32, [4]> x2_5_begin_0 = const()[name = string("x2_5_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_5_end_0 = const()[name = string("x2_5_end_0"), val = tensor<int32, [4]>([1, 32, 64, 64])];
            tensor<bool, [4]> x2_5_end_mask_0 = const()[name = string("x2_5_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 64, 32]> x2_5 = slice_by_index(begin = x2_5_begin_0, end = x2_5_end_0, end_mask = x2_5_end_mask_0, x = x_29)[name = string("x2_5")];
            tensor<fp16, [1, 32, 64, 32]> var_528 = mul(x = x1_5, y = cos_7)[name = string("op_528")];
            tensor<fp16, [1, 32, 64, 32]> var_529 = mul(x = x2_5, y = sin_7)[name = string("op_529")];
            tensor<fp16, [1, 32, 64, 32]> var_530 = sub(x = var_528, y = var_529)[name = string("op_530")];
            tensor<fp16, [1, 32, 64, 32]> var_531 = mul(x = x2_5, y = cos_7)[name = string("op_531")];
            tensor<fp16, [1, 32, 64, 32]> var_532 = mul(x = x1_5, y = sin_7)[name = string("op_532")];
            tensor<fp16, [1, 32, 64, 32]> var_533 = add(x = var_531, y = var_532)[name = string("op_533")];
            bool rotated_5_interleave_0 = const()[name = string("rotated_5_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 32, 64, 64]> rotated_5 = concat(axis = var_44, interleave = rotated_5_interleave_0, values = (var_530, var_533))[name = string("rotated_5")];
            tensor<int32, [4]> x1_7_begin_0 = const()[name = string("x1_7_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_7_end_0 = const()[name = string("x1_7_end_0"), val = tensor<int32, [4]>([1, 8, 64, 32])];
            tensor<bool, [4]> x1_7_end_mask_0 = const()[name = string("x1_7_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 64, 64]> x_33 = transpose(perm = var_506, x = var_505)[name = string("transpose_46")];
            tensor<fp16, [1, 8, 64, 32]> x1_7 = slice_by_index(begin = x1_7_begin_0, end = x1_7_end_0, end_mask = x1_7_end_mask_0, x = x_33)[name = string("x1_7")];
            tensor<int32, [4]> x2_7_begin_0 = const()[name = string("x2_7_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_7_end_0 = const()[name = string("x2_7_end_0"), val = tensor<int32, [4]>([1, 8, 64, 64])];
            tensor<bool, [4]> x2_7_end_mask_0 = const()[name = string("x2_7_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 64, 32]> x2_7 = slice_by_index(begin = x2_7_begin_0, end = x2_7_end_0, end_mask = x2_7_end_mask_0, x = x_33)[name = string("x2_7")];
            tensor<fp16, [1, 8, 64, 32]> var_549 = mul(x = x1_7, y = cos_7)[name = string("op_549")];
            tensor<fp16, [1, 8, 64, 32]> var_550 = mul(x = x2_7, y = sin_7)[name = string("op_550")];
            tensor<fp16, [1, 8, 64, 32]> var_551 = sub(x = var_549, y = var_550)[name = string("op_551")];
            tensor<fp16, [1, 8, 64, 32]> var_552 = mul(x = x2_7, y = cos_7)[name = string("op_552")];
            tensor<fp16, [1, 8, 64, 32]> var_553 = mul(x = x1_7, y = sin_7)[name = string("op_553")];
            tensor<fp16, [1, 8, 64, 32]> var_554 = add(x = var_552, y = var_553)[name = string("op_554")];
            bool rotated_7_interleave_0 = const()[name = string("rotated_7_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 8, 64, 64]> rotated_7 = concat(axis = var_44, interleave = rotated_7_interleave_0, values = (var_551, var_554))[name = string("rotated_7")];
            tensor<int32, [1]> expand_dims_12 = const()[name = string("expand_dims_12"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> expand_dims_13 = const()[name = string("expand_dims_13"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_15 = const()[name = string("expand_dims_15"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_16 = const()[name = string("expand_dims_16"), val = tensor<int32, [1]>([2])];
            int32 concat_20_axis_0 = const()[name = string("concat_20_axis_0"), val = int32(0)];
            bool concat_20_interleave_0 = const()[name = string("concat_20_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_20 = concat(axis = concat_20_axis_0, interleave = concat_20_interleave_0, values = (expand_dims_12, expand_dims_13, current_pos, expand_dims_15))[name = string("concat_20")];
            tensor<int32, [1]> concat_21_values1_0 = const()[name = string("concat_21_values1_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> concat_21_values3_0 = const()[name = string("concat_21_values3_0"), val = tensor<int32, [1]>([0])];
            int32 concat_21_axis_0 = const()[name = string("concat_21_axis_0"), val = int32(0)];
            bool concat_21_interleave_0 = const()[name = string("concat_21_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_21 = concat(axis = concat_21_axis_0, interleave = concat_21_interleave_0, values = (expand_dims_16, concat_21_values1_0, var_366, concat_21_values3_0))[name = string("concat_21")];
            tensor<int32, [4]> model_model_kv_cache_0_internal_tensor_assign_3_stride_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_3_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_3_begin_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_3_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_3_end_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_3_end_mask_0"), val = tensor<bool, [4]>([false, true, false, true])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_3_squeeze_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_3_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<fp16, [32, 8, 2048, 64]> model_model_kv_cache_0_internal_tensor_assign_3_cast_fp16 = slice_update(begin = concat_20, begin_mask = model_model_kv_cache_0_internal_tensor_assign_3_begin_mask_0, end = concat_21, end_mask = model_model_kv_cache_0_internal_tensor_assign_3_end_mask_0, squeeze_mask = model_model_kv_cache_0_internal_tensor_assign_3_squeeze_mask_0, stride = model_model_kv_cache_0_internal_tensor_assign_3_stride_0, update = rotated_7, x = coreml_update_state_17)[name = string("model_model_kv_cache_0_internal_tensor_assign_3_cast_fp16")];
            write_state(data = model_model_kv_cache_0_internal_tensor_assign_3_cast_fp16, input = model_model_kv_cache_0)[name = string("coreml_update_state_18_write_state")];
            tensor<fp16, [32, 8, 2048, 64]> coreml_update_state_18 = read_state(input = model_model_kv_cache_0)[name = string("coreml_update_state_18")];
            tensor<int32, [1]> expand_dims_18 = const()[name = string("expand_dims_18"), val = tensor<int32, [1]>([17])];
            tensor<int32, [1]> expand_dims_19 = const()[name = string("expand_dims_19"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_21 = const()[name = string("expand_dims_21"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_22 = const()[name = string("expand_dims_22"), val = tensor<int32, [1]>([18])];
            int32 concat_24_axis_0 = const()[name = string("concat_24_axis_0"), val = int32(0)];
            bool concat_24_interleave_0 = const()[name = string("concat_24_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_24 = concat(axis = concat_24_axis_0, interleave = concat_24_interleave_0, values = (expand_dims_18, expand_dims_19, current_pos, expand_dims_21))[name = string("concat_24")];
            tensor<int32, [1]> concat_25_values1_0 = const()[name = string("concat_25_values1_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> concat_25_values3_0 = const()[name = string("concat_25_values3_0"), val = tensor<int32, [1]>([0])];
            int32 concat_25_axis_0 = const()[name = string("concat_25_axis_0"), val = int32(0)];
            bool concat_25_interleave_0 = const()[name = string("concat_25_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_25 = concat(axis = concat_25_axis_0, interleave = concat_25_interleave_0, values = (expand_dims_22, concat_25_values1_0, var_366, concat_25_values3_0))[name = string("concat_25")];
            tensor<int32, [4]> model_model_kv_cache_0_internal_tensor_assign_4_stride_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_4_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_4_begin_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_4_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_4_end_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_4_end_mask_0"), val = tensor<bool, [4]>([false, true, false, true])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_4_squeeze_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_4_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<fp16, [1, 8, 64, 64]> value_states_9 = transpose(perm = var_510, x = var_509)[name = string("transpose_45")];
            tensor<fp16, [32, 8, 2048, 64]> model_model_kv_cache_0_internal_tensor_assign_4_cast_fp16 = slice_update(begin = concat_24, begin_mask = model_model_kv_cache_0_internal_tensor_assign_4_begin_mask_0, end = concat_25, end_mask = model_model_kv_cache_0_internal_tensor_assign_4_end_mask_0, squeeze_mask = model_model_kv_cache_0_internal_tensor_assign_4_squeeze_mask_0, stride = model_model_kv_cache_0_internal_tensor_assign_4_stride_0, update = value_states_9, x = coreml_update_state_18)[name = string("model_model_kv_cache_0_internal_tensor_assign_4_cast_fp16")];
            write_state(data = model_model_kv_cache_0_internal_tensor_assign_4_cast_fp16, input = model_model_kv_cache_0)[name = string("coreml_update_state_19_write_state")];
            tensor<fp16, [32, 8, 2048, 64]> coreml_update_state_19 = read_state(input = model_model_kv_cache_0)[name = string("coreml_update_state_19")];
            tensor<int32, [4]> var_577_begin_0 = const()[name = string("op_577_begin_0"), val = tensor<int32, [4]>([1, 0, 0, 0])];
            tensor<int32, [4]> var_577_end_0 = const()[name = string("op_577_end_0"), val = tensor<int32, [4]>([2, 8, 2048, 64])];
            tensor<bool, [4]> var_577_end_mask_0 = const()[name = string("op_577_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<fp16, [1, 8, 2048, 64]> var_577_cast_fp16 = slice_by_index(begin = var_577_begin_0, end = var_577_end_0, end_mask = var_577_end_mask_0, x = coreml_update_state_19)[name = string("op_577_cast_fp16")];
            tensor<int32, [1]> K_layer_cache_3_axes_0 = const()[name = string("K_layer_cache_3_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<fp16, [8, 2048, 64]> K_layer_cache_3_cast_fp16 = squeeze(axes = K_layer_cache_3_axes_0, x = var_577_cast_fp16)[name = string("K_layer_cache_3_cast_fp16")];
            tensor<int32, [4]> var_579_begin_0 = const()[name = string("op_579_begin_0"), val = tensor<int32, [4]>([17, 0, 0, 0])];
            tensor<int32, [4]> var_579_end_0 = const()[name = string("op_579_end_0"), val = tensor<int32, [4]>([18, 8, 2048, 64])];
            tensor<bool, [4]> var_579_end_mask_0 = const()[name = string("op_579_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<fp16, [1, 8, 2048, 64]> var_579_cast_fp16 = slice_by_index(begin = var_579_begin_0, end = var_579_end_0, end_mask = var_579_end_mask_0, x = coreml_update_state_19)[name = string("op_579_cast_fp16")];
            tensor<int32, [1]> V_layer_cache_3_axes_0 = const()[name = string("V_layer_cache_3_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<fp16, [8, 2048, 64]> V_layer_cache_3_cast_fp16 = squeeze(axes = V_layer_cache_3_axes_0, x = var_579_cast_fp16)[name = string("V_layer_cache_3_cast_fp16")];
            tensor<int32, [1]> x_39_axes_0 = const()[name = string("x_39_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [8, 1, 2048, 64]> x_39_cast_fp16 = expand_dims(axes = x_39_axes_0, x = K_layer_cache_3_cast_fp16)[name = string("x_39_cast_fp16")];
            tensor<int32, [4]> var_588 = const()[name = string("op_588"), val = tensor<int32, [4]>([1, 4, 1, 1])];
            tensor<fp16, [8, 4, 2048, 64]> x_41_cast_fp16 = tile(reps = var_588, x = x_39_cast_fp16)[name = string("x_41_cast_fp16")];
            tensor<int32, [4]> var_592 = const()[name = string("op_592"), val = tensor<int32, [4]>([1, -1, 2048, 64])];
            tensor<fp16, [1, 32, 2048, 64]> var_593_cast_fp16 = reshape(shape = var_592, x = x_41_cast_fp16)[name = string("op_593_cast_fp16")];
            tensor<int32, [1]> x_45_axes_0 = const()[name = string("x_45_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [8, 1, 2048, 64]> x_45_cast_fp16 = expand_dims(axes = x_45_axes_0, x = V_layer_cache_3_cast_fp16)[name = string("x_45_cast_fp16")];
            tensor<int32, [4]> var_595 = const()[name = string("op_595"), val = tensor<int32, [4]>([1, 4, 1, 1])];
            tensor<fp16, [8, 4, 2048, 64]> x_47_cast_fp16 = tile(reps = var_595, x = x_45_cast_fp16)[name = string("x_47_cast_fp16")];
            bool var_602_transpose_x_0 = const()[name = string("op_602_transpose_x_0"), val = bool(false)];
            bool var_602_transpose_y_0 = const()[name = string("op_602_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 32, 64, 2048]> var_602_cast_fp16 = matmul(transpose_x = var_602_transpose_x_0, transpose_y = var_602_transpose_y_0, x = rotated_5, y = var_593_cast_fp16)[name = string("op_602_cast_fp16")];
            fp16 var_603_to_fp16 = const()[name = string("op_603_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 32, 64, 2048]> attn_weights_3_cast_fp16 = mul(x = var_602_cast_fp16, y = var_603_to_fp16)[name = string("attn_weights_3_cast_fp16")];
            tensor<fp16, [1, 32, 64, 2048]> x_49_cast_fp16 = add(x = attn_weights_3_cast_fp16, y = causal_mask)[name = string("x_49_cast_fp16")];
            tensor<int32, [1]> reduce_max_1_axes_0 = const()[name = string("reduce_max_1_axes_0"), val = tensor<int32, [1]>([-1])];
            bool reduce_max_1_keep_dims_0 = const()[name = string("reduce_max_1_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 64, 1]> reduce_max_1_cast_fp16 = reduce_max(axes = reduce_max_1_axes_0, keep_dims = reduce_max_1_keep_dims_0, x = x_49_cast_fp16)[name = string("reduce_max_1_cast_fp16")];
            tensor<fp16, [1, 32, 64, 2048]> x_51_cast_fp16 = sub(x = x_49_cast_fp16, y = reduce_max_1_cast_fp16)[name = string("x_51_cast_fp16")];
            tensor<fp16, [1, 32, 64, 2048]> exp_x_3_cast_fp16 = exp(x = x_51_cast_fp16)[name = string("exp_x_3_cast_fp16")];
            tensor<int32, [1]> var_614_axes_0 = const()[name = string("op_614_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_614_keep_dims_0 = const()[name = string("op_614_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 64, 1]> var_614_cast_fp16 = reduce_sum(axes = var_614_axes_0, keep_dims = var_614_keep_dims_0, x = exp_x_3_cast_fp16)[name = string("op_614_cast_fp16")];
            tensor<fp16, [1, 32, 64, 2048]> var_615_cast_fp16 = real_div(x = exp_x_3_cast_fp16, y = var_614_cast_fp16)[name = string("op_615_cast_fp16")];
            tensor<int32, [3]> concat_30 = const()[name = string("concat_30"), val = tensor<int32, [3]>([32, 64, 2048])];
            tensor<fp16, [32, 64, 2048]> reshape_3_cast_fp16 = reshape(shape = concat_30, x = var_615_cast_fp16)[name = string("reshape_3_cast_fp16")];
            tensor<int32, [3]> concat_31 = const()[name = string("concat_31"), val = tensor<int32, [3]>([32, 2048, 64])];
            tensor<fp16, [32, 2048, 64]> reshape_4_cast_fp16 = reshape(shape = concat_31, x = x_47_cast_fp16)[name = string("reshape_4_cast_fp16")];
            bool matmul_1_transpose_x_0 = const()[name = string("matmul_1_transpose_x_0"), val = bool(false)];
            bool matmul_1_transpose_y_0 = const()[name = string("matmul_1_transpose_y_0"), val = bool(false)];
            tensor<fp16, [32, 64, 64]> matmul_1_cast_fp16 = matmul(transpose_x = matmul_1_transpose_x_0, transpose_y = matmul_1_transpose_y_0, x = reshape_3_cast_fp16, y = reshape_4_cast_fp16)[name = string("matmul_1_cast_fp16")];
            tensor<int32, [4]> concat_35 = const()[name = string("concat_35"), val = tensor<int32, [4]>([1, 32, 64, 64])];
            tensor<fp16, [1, 32, 64, 64]> reshape_5_cast_fp16 = reshape(shape = concat_35, x = matmul_1_cast_fp16)[name = string("reshape_5_cast_fp16")];
            tensor<int32, [4]> var_618_perm_0 = const()[name = string("op_618_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_620 = const()[name = string("op_620"), val = tensor<int32, [3]>([1, 64, 2048])];
            tensor<fp16, [1, 64, 32, 64]> var_618_cast_fp16 = transpose(perm = var_618_perm_0, x = reshape_5_cast_fp16)[name = string("transpose_44")];
            tensor<fp16, [1, 64, 2048]> input_19_cast_fp16 = reshape(shape = var_620, x = var_618_cast_fp16)[name = string("input_19_cast_fp16")];
            tensor<fp16, [2048, 2048]> model_model_layers_1_self_attn_o_proj_weight_promoted_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262863424))), lut = tensor<fp16, [256, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(264960640))))[name = string("model_model_layers_1_self_attn_o_proj_weight_promoted_to_fp16_palettized")];
            tensor<fp16, [1, 64, 2048]> linear_1_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_1_self_attn_o_proj_weight_promoted_to_fp16_palettized, x = input_19_cast_fp16)[name = string("linear_1_cast_fp16")];
            tensor<fp16, [1, 64, 2048]> hidden_states_13_cast_fp16 = add(x = hidden_states_9_cast_fp16, y = linear_1_cast_fp16)[name = string("hidden_states_13_cast_fp16")];
            tensor<int32, [1]> mean_7_axes_0 = const()[name = string("mean_7_axes_0"), val = tensor<int32, [1]>([-1])];
            bool mean_7_keep_dims_0 = const()[name = string("mean_7_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 64, 1]> mean_7_cast_fp16 = reduce_mean(axes = mean_7_axes_0, keep_dims = mean_7_keep_dims_0, x = hidden_states_13_cast_fp16)[name = string("mean_7_cast_fp16")];
            tensor<fp16, [1, 64, 2048]> input_21_cast_fp16 = sub(x = hidden_states_13_cast_fp16, y = mean_7_cast_fp16)[name = string("input_21_cast_fp16")];
            tensor<int32, [1]> var_631_axes_0 = const()[name = string("op_631_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2048]> model_model_layers_1_post_attention_layernorm_weight_to_fp16 = const()[name = string("model_model_layers_1_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(264968896)))];
            tensor<fp16, [1, 64, 2048]> var_631_cast_fp16 = layer_norm(axes = var_631_axes_0, epsilon = var_46_to_fp16, gamma = model_model_layers_1_post_attention_layernorm_weight_to_fp16, x = input_21_cast_fp16)[name = string("op_631_cast_fp16")];
            tensor<int32, [3]> var_638 = const()[name = string("op_638"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> input_23_axes_0 = const()[name = string("input_23_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 64]> var_639 = transpose(perm = var_638, x = var_631_cast_fp16)[name = string("transpose_43")];
            tensor<fp16, [1, 2048, 1, 64]> input_23 = expand_dims(axes = input_23_axes_0, x = var_639)[name = string("input_23")];
            string input_25_pad_type_0 = const()[name = string("input_25_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_25_strides_0 = const()[name = string("input_25_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_25_pad_0 = const()[name = string("input_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_25_dilations_0 = const()[name = string("input_25_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_25_groups_0 = const()[name = string("input_25_groups_0"), val = int32(1)];
            tensor<fp16, [1, 8192, 1, 64]> input_25 = conv(dilations = input_25_dilations_0, groups = input_25_groups_0, pad = input_25_pad_0, pad_type = input_25_pad_type_0, strides = input_25_strides_0, weight = model_model_layers_1_mlp_gate_proj_weight_palettized, x = input_23)[name = string("input_25")];
            string up_states_3_pad_type_0 = const()[name = string("up_states_3_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> up_states_3_strides_0 = const()[name = string("up_states_3_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> up_states_3_pad_0 = const()[name = string("up_states_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> up_states_3_dilations_0 = const()[name = string("up_states_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 up_states_3_groups_0 = const()[name = string("up_states_3_groups_0"), val = int32(1)];
            tensor<fp16, [1, 8192, 1, 64]> up_states_3 = conv(dilations = up_states_3_dilations_0, groups = up_states_3_groups_0, pad = up_states_3_pad_0, pad_type = up_states_3_pad_type_0, strides = up_states_3_strides_0, weight = model_model_layers_1_mlp_up_proj_weight_palettized, x = input_23)[name = string("up_states_3")];
            tensor<fp16, [1, 8192, 1, 64]> gate_states_3 = silu(x = input_25)[name = string("gate_states_3")];
            tensor<fp16, [1, 8192, 1, 64]> input_27 = mul(x = gate_states_3, y = up_states_3)[name = string("input_27")];
            string hidden_states_15_pad_type_0 = const()[name = string("hidden_states_15_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_15_strides_0 = const()[name = string("hidden_states_15_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_15_pad_0 = const()[name = string("hidden_states_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_15_dilations_0 = const()[name = string("hidden_states_15_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_15_groups_0 = const()[name = string("hidden_states_15_groups_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 1, 64]> hidden_states_15 = conv(dilations = hidden_states_15_dilations_0, groups = hidden_states_15_groups_0, pad = hidden_states_15_pad_0, pad_type = hidden_states_15_pad_type_0, strides = hidden_states_15_strides_0, weight = model_model_layers_1_mlp_down_proj_weight_palettized, x = input_27)[name = string("hidden_states_15")];
            tensor<int32, [1]> var_661_axes_0 = const()[name = string("op_661_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 64]> var_661 = squeeze(axes = var_661_axes_0, x = hidden_states_15)[name = string("op_661")];
            tensor<int32, [3]> var_662 = const()[name = string("op_662"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 64, 2048]> var_663 = transpose(perm = var_662, x = var_661)[name = string("transpose_42")];
            tensor<fp16, [1, 64, 2048]> hidden_states_17_cast_fp16 = add(x = hidden_states_13_cast_fp16, y = var_663)[name = string("hidden_states_17_cast_fp16")];
            tensor<int32, [1]> mean_9_axes_0 = const()[name = string("mean_9_axes_0"), val = tensor<int32, [1]>([-1])];
            bool mean_9_keep_dims_0 = const()[name = string("mean_9_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 64, 1]> mean_9_cast_fp16 = reduce_mean(axes = mean_9_axes_0, keep_dims = mean_9_keep_dims_0, x = hidden_states_17_cast_fp16)[name = string("mean_9_cast_fp16")];
            tensor<fp16, [1, 64, 2048]> input_29_cast_fp16 = sub(x = hidden_states_17_cast_fp16, y = mean_9_cast_fp16)[name = string("input_29_cast_fp16")];
            tensor<int32, [1]> var_671_axes_0 = const()[name = string("op_671_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2048]> model_model_layers_2_input_layernorm_weight_to_fp16 = const()[name = string("model_model_layers_2_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(264973056)))];
            tensor<fp16, [1, 64, 2048]> var_671_cast_fp16 = layer_norm(axes = var_671_axes_0, epsilon = var_46_to_fp16, gamma = model_model_layers_2_input_layernorm_weight_to_fp16, x = input_29_cast_fp16)[name = string("op_671_cast_fp16")];
            tensor<int32, [3]> var_675 = const()[name = string("op_675"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_677_axes_0 = const()[name = string("op_677_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 64]> var_676 = transpose(perm = var_675, x = var_671_cast_fp16)[name = string("transpose_41")];
            tensor<fp16, [1, 2048, 1, 64]> var_677 = expand_dims(axes = var_677_axes_0, x = var_676)[name = string("op_677")];
            string query_states_9_pad_type_0 = const()[name = string("query_states_9_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_states_9_strides_0 = const()[name = string("query_states_9_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_states_9_pad_0 = const()[name = string("query_states_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_states_9_dilations_0 = const()[name = string("query_states_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_states_9_groups_0 = const()[name = string("query_states_9_groups_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 1, 64]> query_states_9 = conv(dilations = query_states_9_dilations_0, groups = query_states_9_groups_0, pad = query_states_9_pad_0, pad_type = query_states_9_pad_type_0, strides = query_states_9_strides_0, weight = model_model_layers_2_self_attn_q_proj_weight_palettized, x = var_677)[name = string("query_states_9")];
            string key_states_13_pad_type_0 = const()[name = string("key_states_13_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_states_13_strides_0 = const()[name = string("key_states_13_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_states_13_pad_0 = const()[name = string("key_states_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_states_13_dilations_0 = const()[name = string("key_states_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_states_13_groups_0 = const()[name = string("key_states_13_groups_0"), val = int32(1)];
            tensor<fp16, [1, 512, 1, 64]> key_states_13 = conv(dilations = key_states_13_dilations_0, groups = key_states_13_groups_0, pad = key_states_13_pad_0, pad_type = key_states_13_pad_type_0, strides = key_states_13_strides_0, weight = model_model_layers_2_self_attn_k_proj_weight_palettized, x = var_677)[name = string("key_states_13")];
            string value_states_13_pad_type_0 = const()[name = string("value_states_13_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_states_13_strides_0 = const()[name = string("value_states_13_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_states_13_pad_0 = const()[name = string("value_states_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_states_13_dilations_0 = const()[name = string("value_states_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_states_13_groups_0 = const()[name = string("value_states_13_groups_0"), val = int32(1)];
            tensor<fp16, [1, 512, 1, 64]> value_states_13 = conv(dilations = value_states_13_dilations_0, groups = value_states_13_groups_0, pad = value_states_13_pad_0, pad_type = value_states_13_pad_type_0, strides = value_states_13_strides_0, weight = model_model_layers_2_self_attn_v_proj_weight_palettized, x = var_677)[name = string("value_states_13")];
            tensor<int32, [4]> var_697 = const()[name = string("op_697"), val = tensor<int32, [4]>([1, 32, 64, 64])];
            tensor<fp16, [1, 32, 64, 64]> var_698 = reshape(shape = var_697, x = query_states_9)[name = string("op_698")];
            tensor<int32, [4]> var_699 = const()[name = string("op_699"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_701 = const()[name = string("op_701"), val = tensor<int32, [4]>([1, 8, 64, 64])];
            tensor<fp16, [1, 8, 64, 64]> var_702 = reshape(shape = var_701, x = key_states_13)[name = string("op_702")];
            tensor<int32, [4]> var_703 = const()[name = string("op_703"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_705 = const()[name = string("op_705"), val = tensor<int32, [4]>([1, 8, 64, 64])];
            tensor<fp16, [1, 8, 64, 64]> var_706 = reshape(shape = var_705, x = value_states_13)[name = string("op_706")];
            tensor<int32, [4]> var_707 = const()[name = string("op_707"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> x1_9_begin_0 = const()[name = string("x1_9_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_9_end_0 = const()[name = string("x1_9_end_0"), val = tensor<int32, [4]>([1, 32, 64, 32])];
            tensor<bool, [4]> x1_9_end_mask_0 = const()[name = string("x1_9_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 64, 64]> x_57 = transpose(perm = var_699, x = var_698)[name = string("transpose_40")];
            tensor<fp16, [1, 32, 64, 32]> x1_9 = slice_by_index(begin = x1_9_begin_0, end = x1_9_end_0, end_mask = x1_9_end_mask_0, x = x_57)[name = string("x1_9")];
            tensor<int32, [4]> x2_9_begin_0 = const()[name = string("x2_9_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_9_end_0 = const()[name = string("x2_9_end_0"), val = tensor<int32, [4]>([1, 32, 64, 64])];
            tensor<bool, [4]> x2_9_end_mask_0 = const()[name = string("x2_9_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 64, 32]> x2_9 = slice_by_index(begin = x2_9_begin_0, end = x2_9_end_0, end_mask = x2_9_end_mask_0, x = x_57)[name = string("x2_9")];
            tensor<fp16, [1, 32, 64, 32]> var_725 = mul(x = x1_9, y = cos_7)[name = string("op_725")];
            tensor<fp16, [1, 32, 64, 32]> var_726 = mul(x = x2_9, y = sin_7)[name = string("op_726")];
            tensor<fp16, [1, 32, 64, 32]> var_727 = sub(x = var_725, y = var_726)[name = string("op_727")];
            tensor<fp16, [1, 32, 64, 32]> var_728 = mul(x = x2_9, y = cos_7)[name = string("op_728")];
            tensor<fp16, [1, 32, 64, 32]> var_729 = mul(x = x1_9, y = sin_7)[name = string("op_729")];
            tensor<fp16, [1, 32, 64, 32]> var_730 = add(x = var_728, y = var_729)[name = string("op_730")];
            bool rotated_9_interleave_0 = const()[name = string("rotated_9_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 32, 64, 64]> rotated_9 = concat(axis = var_44, interleave = rotated_9_interleave_0, values = (var_727, var_730))[name = string("rotated_9")];
            tensor<int32, [4]> x1_11_begin_0 = const()[name = string("x1_11_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_11_end_0 = const()[name = string("x1_11_end_0"), val = tensor<int32, [4]>([1, 8, 64, 32])];
            tensor<bool, [4]> x1_11_end_mask_0 = const()[name = string("x1_11_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 64, 64]> x_61 = transpose(perm = var_703, x = var_702)[name = string("transpose_39")];
            tensor<fp16, [1, 8, 64, 32]> x1_11 = slice_by_index(begin = x1_11_begin_0, end = x1_11_end_0, end_mask = x1_11_end_mask_0, x = x_61)[name = string("x1_11")];
            tensor<int32, [4]> x2_11_begin_0 = const()[name = string("x2_11_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_11_end_0 = const()[name = string("x2_11_end_0"), val = tensor<int32, [4]>([1, 8, 64, 64])];
            tensor<bool, [4]> x2_11_end_mask_0 = const()[name = string("x2_11_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 64, 32]> x2_11 = slice_by_index(begin = x2_11_begin_0, end = x2_11_end_0, end_mask = x2_11_end_mask_0, x = x_61)[name = string("x2_11")];
            tensor<fp16, [1, 8, 64, 32]> var_746 = mul(x = x1_11, y = cos_7)[name = string("op_746")];
            tensor<fp16, [1, 8, 64, 32]> var_747 = mul(x = x2_11, y = sin_7)[name = string("op_747")];
            tensor<fp16, [1, 8, 64, 32]> var_748 = sub(x = var_746, y = var_747)[name = string("op_748")];
            tensor<fp16, [1, 8, 64, 32]> var_749 = mul(x = x2_11, y = cos_7)[name = string("op_749")];
            tensor<fp16, [1, 8, 64, 32]> var_750 = mul(x = x1_11, y = sin_7)[name = string("op_750")];
            tensor<fp16, [1, 8, 64, 32]> var_751 = add(x = var_749, y = var_750)[name = string("op_751")];
            bool rotated_11_interleave_0 = const()[name = string("rotated_11_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 8, 64, 64]> rotated_11 = concat(axis = var_44, interleave = rotated_11_interleave_0, values = (var_748, var_751))[name = string("rotated_11")];
            tensor<int32, [1]> expand_dims_24 = const()[name = string("expand_dims_24"), val = tensor<int32, [1]>([2])];
            tensor<int32, [1]> expand_dims_25 = const()[name = string("expand_dims_25"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_27 = const()[name = string("expand_dims_27"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_28 = const()[name = string("expand_dims_28"), val = tensor<int32, [1]>([3])];
            int32 concat_38_axis_0 = const()[name = string("concat_38_axis_0"), val = int32(0)];
            bool concat_38_interleave_0 = const()[name = string("concat_38_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_38 = concat(axis = concat_38_axis_0, interleave = concat_38_interleave_0, values = (expand_dims_24, expand_dims_25, current_pos, expand_dims_27))[name = string("concat_38")];
            tensor<int32, [1]> concat_39_values1_0 = const()[name = string("concat_39_values1_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> concat_39_values3_0 = const()[name = string("concat_39_values3_0"), val = tensor<int32, [1]>([0])];
            int32 concat_39_axis_0 = const()[name = string("concat_39_axis_0"), val = int32(0)];
            bool concat_39_interleave_0 = const()[name = string("concat_39_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_39 = concat(axis = concat_39_axis_0, interleave = concat_39_interleave_0, values = (expand_dims_28, concat_39_values1_0, var_366, concat_39_values3_0))[name = string("concat_39")];
            tensor<int32, [4]> model_model_kv_cache_0_internal_tensor_assign_5_stride_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_5_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_5_begin_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_5_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_5_end_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_5_end_mask_0"), val = tensor<bool, [4]>([false, true, false, true])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_5_squeeze_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_5_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<fp16, [32, 8, 2048, 64]> model_model_kv_cache_0_internal_tensor_assign_5_cast_fp16 = slice_update(begin = concat_38, begin_mask = model_model_kv_cache_0_internal_tensor_assign_5_begin_mask_0, end = concat_39, end_mask = model_model_kv_cache_0_internal_tensor_assign_5_end_mask_0, squeeze_mask = model_model_kv_cache_0_internal_tensor_assign_5_squeeze_mask_0, stride = model_model_kv_cache_0_internal_tensor_assign_5_stride_0, update = rotated_11, x = coreml_update_state_19)[name = string("model_model_kv_cache_0_internal_tensor_assign_5_cast_fp16")];
            write_state(data = model_model_kv_cache_0_internal_tensor_assign_5_cast_fp16, input = model_model_kv_cache_0)[name = string("coreml_update_state_20_write_state")];
            tensor<fp16, [32, 8, 2048, 64]> coreml_update_state_20 = read_state(input = model_model_kv_cache_0)[name = string("coreml_update_state_20")];
            tensor<int32, [1]> expand_dims_30 = const()[name = string("expand_dims_30"), val = tensor<int32, [1]>([18])];
            tensor<int32, [1]> expand_dims_31 = const()[name = string("expand_dims_31"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_33 = const()[name = string("expand_dims_33"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_34 = const()[name = string("expand_dims_34"), val = tensor<int32, [1]>([19])];
            int32 concat_42_axis_0 = const()[name = string("concat_42_axis_0"), val = int32(0)];
            bool concat_42_interleave_0 = const()[name = string("concat_42_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_42 = concat(axis = concat_42_axis_0, interleave = concat_42_interleave_0, values = (expand_dims_30, expand_dims_31, current_pos, expand_dims_33))[name = string("concat_42")];
            tensor<int32, [1]> concat_43_values1_0 = const()[name = string("concat_43_values1_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> concat_43_values3_0 = const()[name = string("concat_43_values3_0"), val = tensor<int32, [1]>([0])];
            int32 concat_43_axis_0 = const()[name = string("concat_43_axis_0"), val = int32(0)];
            bool concat_43_interleave_0 = const()[name = string("concat_43_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_43 = concat(axis = concat_43_axis_0, interleave = concat_43_interleave_0, values = (expand_dims_34, concat_43_values1_0, var_366, concat_43_values3_0))[name = string("concat_43")];
            tensor<int32, [4]> model_model_kv_cache_0_internal_tensor_assign_6_stride_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_6_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_6_begin_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_6_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_6_end_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_6_end_mask_0"), val = tensor<bool, [4]>([false, true, false, true])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_6_squeeze_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_6_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<fp16, [1, 8, 64, 64]> value_states_15 = transpose(perm = var_707, x = var_706)[name = string("transpose_38")];
            tensor<fp16, [32, 8, 2048, 64]> model_model_kv_cache_0_internal_tensor_assign_6_cast_fp16 = slice_update(begin = concat_42, begin_mask = model_model_kv_cache_0_internal_tensor_assign_6_begin_mask_0, end = concat_43, end_mask = model_model_kv_cache_0_internal_tensor_assign_6_end_mask_0, squeeze_mask = model_model_kv_cache_0_internal_tensor_assign_6_squeeze_mask_0, stride = model_model_kv_cache_0_internal_tensor_assign_6_stride_0, update = value_states_15, x = coreml_update_state_20)[name = string("model_model_kv_cache_0_internal_tensor_assign_6_cast_fp16")];
            write_state(data = model_model_kv_cache_0_internal_tensor_assign_6_cast_fp16, input = model_model_kv_cache_0)[name = string("coreml_update_state_21_write_state")];
            tensor<fp16, [32, 8, 2048, 64]> coreml_update_state_21 = read_state(input = model_model_kv_cache_0)[name = string("coreml_update_state_21")];
            tensor<int32, [4]> var_774_begin_0 = const()[name = string("op_774_begin_0"), val = tensor<int32, [4]>([2, 0, 0, 0])];
            tensor<int32, [4]> var_774_end_0 = const()[name = string("op_774_end_0"), val = tensor<int32, [4]>([3, 8, 2048, 64])];
            tensor<bool, [4]> var_774_end_mask_0 = const()[name = string("op_774_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<fp16, [1, 8, 2048, 64]> var_774_cast_fp16 = slice_by_index(begin = var_774_begin_0, end = var_774_end_0, end_mask = var_774_end_mask_0, x = coreml_update_state_21)[name = string("op_774_cast_fp16")];
            tensor<int32, [1]> K_layer_cache_5_axes_0 = const()[name = string("K_layer_cache_5_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<fp16, [8, 2048, 64]> K_layer_cache_5_cast_fp16 = squeeze(axes = K_layer_cache_5_axes_0, x = var_774_cast_fp16)[name = string("K_layer_cache_5_cast_fp16")];
            tensor<int32, [4]> var_776_begin_0 = const()[name = string("op_776_begin_0"), val = tensor<int32, [4]>([18, 0, 0, 0])];
            tensor<int32, [4]> var_776_end_0 = const()[name = string("op_776_end_0"), val = tensor<int32, [4]>([19, 8, 2048, 64])];
            tensor<bool, [4]> var_776_end_mask_0 = const()[name = string("op_776_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<fp16, [1, 8, 2048, 64]> var_776_cast_fp16 = slice_by_index(begin = var_776_begin_0, end = var_776_end_0, end_mask = var_776_end_mask_0, x = coreml_update_state_21)[name = string("op_776_cast_fp16")];
            tensor<int32, [1]> V_layer_cache_5_axes_0 = const()[name = string("V_layer_cache_5_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<fp16, [8, 2048, 64]> V_layer_cache_5_cast_fp16 = squeeze(axes = V_layer_cache_5_axes_0, x = var_776_cast_fp16)[name = string("V_layer_cache_5_cast_fp16")];
            tensor<int32, [1]> x_67_axes_0 = const()[name = string("x_67_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [8, 1, 2048, 64]> x_67_cast_fp16 = expand_dims(axes = x_67_axes_0, x = K_layer_cache_5_cast_fp16)[name = string("x_67_cast_fp16")];
            tensor<int32, [4]> var_785 = const()[name = string("op_785"), val = tensor<int32, [4]>([1, 4, 1, 1])];
            tensor<fp16, [8, 4, 2048, 64]> x_69_cast_fp16 = tile(reps = var_785, x = x_67_cast_fp16)[name = string("x_69_cast_fp16")];
            tensor<int32, [4]> var_789 = const()[name = string("op_789"), val = tensor<int32, [4]>([1, -1, 2048, 64])];
            tensor<fp16, [1, 32, 2048, 64]> var_790_cast_fp16 = reshape(shape = var_789, x = x_69_cast_fp16)[name = string("op_790_cast_fp16")];
            tensor<int32, [1]> x_73_axes_0 = const()[name = string("x_73_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [8, 1, 2048, 64]> x_73_cast_fp16 = expand_dims(axes = x_73_axes_0, x = V_layer_cache_5_cast_fp16)[name = string("x_73_cast_fp16")];
            tensor<int32, [4]> var_792 = const()[name = string("op_792"), val = tensor<int32, [4]>([1, 4, 1, 1])];
            tensor<fp16, [8, 4, 2048, 64]> x_75_cast_fp16 = tile(reps = var_792, x = x_73_cast_fp16)[name = string("x_75_cast_fp16")];
            bool var_799_transpose_x_0 = const()[name = string("op_799_transpose_x_0"), val = bool(false)];
            bool var_799_transpose_y_0 = const()[name = string("op_799_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 32, 64, 2048]> var_799_cast_fp16 = matmul(transpose_x = var_799_transpose_x_0, transpose_y = var_799_transpose_y_0, x = rotated_9, y = var_790_cast_fp16)[name = string("op_799_cast_fp16")];
            fp16 var_800_to_fp16 = const()[name = string("op_800_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 32, 64, 2048]> attn_weights_5_cast_fp16 = mul(x = var_799_cast_fp16, y = var_800_to_fp16)[name = string("attn_weights_5_cast_fp16")];
            tensor<fp16, [1, 32, 64, 2048]> x_77_cast_fp16 = add(x = attn_weights_5_cast_fp16, y = causal_mask)[name = string("x_77_cast_fp16")];
            tensor<int32, [1]> reduce_max_2_axes_0 = const()[name = string("reduce_max_2_axes_0"), val = tensor<int32, [1]>([-1])];
            bool reduce_max_2_keep_dims_0 = const()[name = string("reduce_max_2_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 64, 1]> reduce_max_2_cast_fp16 = reduce_max(axes = reduce_max_2_axes_0, keep_dims = reduce_max_2_keep_dims_0, x = x_77_cast_fp16)[name = string("reduce_max_2_cast_fp16")];
            tensor<fp16, [1, 32, 64, 2048]> x_79_cast_fp16 = sub(x = x_77_cast_fp16, y = reduce_max_2_cast_fp16)[name = string("x_79_cast_fp16")];
            tensor<fp16, [1, 32, 64, 2048]> exp_x_5_cast_fp16 = exp(x = x_79_cast_fp16)[name = string("exp_x_5_cast_fp16")];
            tensor<int32, [1]> var_811_axes_0 = const()[name = string("op_811_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_811_keep_dims_0 = const()[name = string("op_811_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 64, 1]> var_811_cast_fp16 = reduce_sum(axes = var_811_axes_0, keep_dims = var_811_keep_dims_0, x = exp_x_5_cast_fp16)[name = string("op_811_cast_fp16")];
            tensor<fp16, [1, 32, 64, 2048]> var_812_cast_fp16 = real_div(x = exp_x_5_cast_fp16, y = var_811_cast_fp16)[name = string("op_812_cast_fp16")];
            tensor<int32, [3]> concat_48 = const()[name = string("concat_48"), val = tensor<int32, [3]>([32, 64, 2048])];
            tensor<fp16, [32, 64, 2048]> reshape_6_cast_fp16 = reshape(shape = concat_48, x = var_812_cast_fp16)[name = string("reshape_6_cast_fp16")];
            tensor<int32, [3]> concat_49 = const()[name = string("concat_49"), val = tensor<int32, [3]>([32, 2048, 64])];
            tensor<fp16, [32, 2048, 64]> reshape_7_cast_fp16 = reshape(shape = concat_49, x = x_75_cast_fp16)[name = string("reshape_7_cast_fp16")];
            bool matmul_2_transpose_x_0 = const()[name = string("matmul_2_transpose_x_0"), val = bool(false)];
            bool matmul_2_transpose_y_0 = const()[name = string("matmul_2_transpose_y_0"), val = bool(false)];
            tensor<fp16, [32, 64, 64]> matmul_2_cast_fp16 = matmul(transpose_x = matmul_2_transpose_x_0, transpose_y = matmul_2_transpose_y_0, x = reshape_6_cast_fp16, y = reshape_7_cast_fp16)[name = string("matmul_2_cast_fp16")];
            tensor<int32, [4]> concat_53 = const()[name = string("concat_53"), val = tensor<int32, [4]>([1, 32, 64, 64])];
            tensor<fp16, [1, 32, 64, 64]> reshape_8_cast_fp16 = reshape(shape = concat_53, x = matmul_2_cast_fp16)[name = string("reshape_8_cast_fp16")];
            tensor<int32, [4]> var_815_perm_0 = const()[name = string("op_815_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_817 = const()[name = string("op_817"), val = tensor<int32, [3]>([1, 64, 2048])];
            tensor<fp16, [1, 64, 32, 64]> var_815_cast_fp16 = transpose(perm = var_815_perm_0, x = reshape_8_cast_fp16)[name = string("transpose_37")];
            tensor<fp16, [1, 64, 2048]> input_33_cast_fp16 = reshape(shape = var_817, x = var_815_cast_fp16)[name = string("input_33_cast_fp16")];
            tensor<fp16, [2048, 2048]> model_model_layers_2_self_attn_o_proj_weight_promoted_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(264977216))), lut = tensor<fp16, [256, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267074432))))[name = string("model_model_layers_2_self_attn_o_proj_weight_promoted_to_fp16_palettized")];
            tensor<fp16, [1, 64, 2048]> linear_2_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_2_self_attn_o_proj_weight_promoted_to_fp16_palettized, x = input_33_cast_fp16)[name = string("linear_2_cast_fp16")];
            tensor<fp16, [1, 64, 2048]> hidden_states_21_cast_fp16 = add(x = hidden_states_17_cast_fp16, y = linear_2_cast_fp16)[name = string("hidden_states_21_cast_fp16")];
            tensor<int32, [1]> mean_11_axes_0 = const()[name = string("mean_11_axes_0"), val = tensor<int32, [1]>([-1])];
            bool mean_11_keep_dims_0 = const()[name = string("mean_11_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 64, 1]> mean_11_cast_fp16 = reduce_mean(axes = mean_11_axes_0, keep_dims = mean_11_keep_dims_0, x = hidden_states_21_cast_fp16)[name = string("mean_11_cast_fp16")];
            tensor<fp16, [1, 64, 2048]> input_35_cast_fp16 = sub(x = hidden_states_21_cast_fp16, y = mean_11_cast_fp16)[name = string("input_35_cast_fp16")];
            tensor<int32, [1]> var_828_axes_0 = const()[name = string("op_828_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2048]> model_model_layers_2_post_attention_layernorm_weight_to_fp16 = const()[name = string("model_model_layers_2_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267082688)))];
            tensor<fp16, [1, 64, 2048]> var_828_cast_fp16 = layer_norm(axes = var_828_axes_0, epsilon = var_46_to_fp16, gamma = model_model_layers_2_post_attention_layernorm_weight_to_fp16, x = input_35_cast_fp16)[name = string("op_828_cast_fp16")];
            tensor<int32, [3]> var_835 = const()[name = string("op_835"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> input_37_axes_0 = const()[name = string("input_37_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 64]> var_836 = transpose(perm = var_835, x = var_828_cast_fp16)[name = string("transpose_36")];
            tensor<fp16, [1, 2048, 1, 64]> input_37 = expand_dims(axes = input_37_axes_0, x = var_836)[name = string("input_37")];
            string input_39_pad_type_0 = const()[name = string("input_39_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_39_strides_0 = const()[name = string("input_39_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_39_pad_0 = const()[name = string("input_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_39_dilations_0 = const()[name = string("input_39_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_39_groups_0 = const()[name = string("input_39_groups_0"), val = int32(1)];
            tensor<fp16, [1, 8192, 1, 64]> input_39 = conv(dilations = input_39_dilations_0, groups = input_39_groups_0, pad = input_39_pad_0, pad_type = input_39_pad_type_0, strides = input_39_strides_0, weight = model_model_layers_2_mlp_gate_proj_weight_palettized, x = input_37)[name = string("input_39")];
            string up_states_5_pad_type_0 = const()[name = string("up_states_5_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> up_states_5_strides_0 = const()[name = string("up_states_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> up_states_5_pad_0 = const()[name = string("up_states_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> up_states_5_dilations_0 = const()[name = string("up_states_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 up_states_5_groups_0 = const()[name = string("up_states_5_groups_0"), val = int32(1)];
            tensor<fp16, [1, 8192, 1, 64]> up_states_5 = conv(dilations = up_states_5_dilations_0, groups = up_states_5_groups_0, pad = up_states_5_pad_0, pad_type = up_states_5_pad_type_0, strides = up_states_5_strides_0, weight = model_model_layers_2_mlp_up_proj_weight_palettized, x = input_37)[name = string("up_states_5")];
            tensor<fp16, [1, 8192, 1, 64]> gate_states_5 = silu(x = input_39)[name = string("gate_states_5")];
            tensor<fp16, [1, 8192, 1, 64]> input_41 = mul(x = gate_states_5, y = up_states_5)[name = string("input_41")];
            string hidden_states_23_pad_type_0 = const()[name = string("hidden_states_23_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_23_strides_0 = const()[name = string("hidden_states_23_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_23_pad_0 = const()[name = string("hidden_states_23_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_23_dilations_0 = const()[name = string("hidden_states_23_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_23_groups_0 = const()[name = string("hidden_states_23_groups_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 1, 64]> hidden_states_23 = conv(dilations = hidden_states_23_dilations_0, groups = hidden_states_23_groups_0, pad = hidden_states_23_pad_0, pad_type = hidden_states_23_pad_type_0, strides = hidden_states_23_strides_0, weight = model_model_layers_2_mlp_down_proj_weight_palettized, x = input_41)[name = string("hidden_states_23")];
            tensor<int32, [1]> var_858_axes_0 = const()[name = string("op_858_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 64]> var_858 = squeeze(axes = var_858_axes_0, x = hidden_states_23)[name = string("op_858")];
            tensor<int32, [3]> var_859 = const()[name = string("op_859"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 64, 2048]> var_860 = transpose(perm = var_859, x = var_858)[name = string("transpose_35")];
            tensor<fp16, [1, 64, 2048]> hidden_states_25_cast_fp16 = add(x = hidden_states_21_cast_fp16, y = var_860)[name = string("hidden_states_25_cast_fp16")];
            tensor<int32, [1]> mean_13_axes_0 = const()[name = string("mean_13_axes_0"), val = tensor<int32, [1]>([-1])];
            bool mean_13_keep_dims_0 = const()[name = string("mean_13_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 64, 1]> mean_13_cast_fp16 = reduce_mean(axes = mean_13_axes_0, keep_dims = mean_13_keep_dims_0, x = hidden_states_25_cast_fp16)[name = string("mean_13_cast_fp16")];
            tensor<fp16, [1, 64, 2048]> input_43_cast_fp16 = sub(x = hidden_states_25_cast_fp16, y = mean_13_cast_fp16)[name = string("input_43_cast_fp16")];
            tensor<int32, [1]> var_868_axes_0 = const()[name = string("op_868_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2048]> model_model_layers_3_input_layernorm_weight_to_fp16 = const()[name = string("model_model_layers_3_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267086848)))];
            tensor<fp16, [1, 64, 2048]> var_868_cast_fp16 = layer_norm(axes = var_868_axes_0, epsilon = var_46_to_fp16, gamma = model_model_layers_3_input_layernorm_weight_to_fp16, x = input_43_cast_fp16)[name = string("op_868_cast_fp16")];
            tensor<int32, [3]> var_872 = const()[name = string("op_872"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_874_axes_0 = const()[name = string("op_874_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 64]> var_873 = transpose(perm = var_872, x = var_868_cast_fp16)[name = string("transpose_34")];
            tensor<fp16, [1, 2048, 1, 64]> var_874 = expand_dims(axes = var_874_axes_0, x = var_873)[name = string("op_874")];
            string query_states_13_pad_type_0 = const()[name = string("query_states_13_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_states_13_strides_0 = const()[name = string("query_states_13_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_states_13_pad_0 = const()[name = string("query_states_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_states_13_dilations_0 = const()[name = string("query_states_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_states_13_groups_0 = const()[name = string("query_states_13_groups_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 1, 64]> query_states_13 = conv(dilations = query_states_13_dilations_0, groups = query_states_13_groups_0, pad = query_states_13_pad_0, pad_type = query_states_13_pad_type_0, strides = query_states_13_strides_0, weight = model_model_layers_3_self_attn_q_proj_weight_palettized, x = var_874)[name = string("query_states_13")];
            string key_states_19_pad_type_0 = const()[name = string("key_states_19_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_states_19_strides_0 = const()[name = string("key_states_19_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_states_19_pad_0 = const()[name = string("key_states_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_states_19_dilations_0 = const()[name = string("key_states_19_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_states_19_groups_0 = const()[name = string("key_states_19_groups_0"), val = int32(1)];
            tensor<fp16, [1, 512, 1, 64]> key_states_19 = conv(dilations = key_states_19_dilations_0, groups = key_states_19_groups_0, pad = key_states_19_pad_0, pad_type = key_states_19_pad_type_0, strides = key_states_19_strides_0, weight = model_model_layers_3_self_attn_k_proj_weight_palettized, x = var_874)[name = string("key_states_19")];
            string value_states_19_pad_type_0 = const()[name = string("value_states_19_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_states_19_strides_0 = const()[name = string("value_states_19_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_states_19_pad_0 = const()[name = string("value_states_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_states_19_dilations_0 = const()[name = string("value_states_19_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_states_19_groups_0 = const()[name = string("value_states_19_groups_0"), val = int32(1)];
            tensor<fp16, [1, 512, 1, 64]> value_states_19 = conv(dilations = value_states_19_dilations_0, groups = value_states_19_groups_0, pad = value_states_19_pad_0, pad_type = value_states_19_pad_type_0, strides = value_states_19_strides_0, weight = model_model_layers_3_self_attn_v_proj_weight_palettized, x = var_874)[name = string("value_states_19")];
            tensor<int32, [4]> var_894 = const()[name = string("op_894"), val = tensor<int32, [4]>([1, 32, 64, 64])];
            tensor<fp16, [1, 32, 64, 64]> var_895 = reshape(shape = var_894, x = query_states_13)[name = string("op_895")];
            tensor<int32, [4]> var_896 = const()[name = string("op_896"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_898 = const()[name = string("op_898"), val = tensor<int32, [4]>([1, 8, 64, 64])];
            tensor<fp16, [1, 8, 64, 64]> var_899 = reshape(shape = var_898, x = key_states_19)[name = string("op_899")];
            tensor<int32, [4]> var_900 = const()[name = string("op_900"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_902 = const()[name = string("op_902"), val = tensor<int32, [4]>([1, 8, 64, 64])];
            tensor<fp16, [1, 8, 64, 64]> var_903 = reshape(shape = var_902, x = value_states_19)[name = string("op_903")];
            tensor<int32, [4]> var_904 = const()[name = string("op_904"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> x1_13_begin_0 = const()[name = string("x1_13_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_13_end_0 = const()[name = string("x1_13_end_0"), val = tensor<int32, [4]>([1, 32, 64, 32])];
            tensor<bool, [4]> x1_13_end_mask_0 = const()[name = string("x1_13_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 64, 64]> x_85 = transpose(perm = var_896, x = var_895)[name = string("transpose_33")];
            tensor<fp16, [1, 32, 64, 32]> x1_13 = slice_by_index(begin = x1_13_begin_0, end = x1_13_end_0, end_mask = x1_13_end_mask_0, x = x_85)[name = string("x1_13")];
            tensor<int32, [4]> x2_13_begin_0 = const()[name = string("x2_13_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_13_end_0 = const()[name = string("x2_13_end_0"), val = tensor<int32, [4]>([1, 32, 64, 64])];
            tensor<bool, [4]> x2_13_end_mask_0 = const()[name = string("x2_13_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 64, 32]> x2_13 = slice_by_index(begin = x2_13_begin_0, end = x2_13_end_0, end_mask = x2_13_end_mask_0, x = x_85)[name = string("x2_13")];
            tensor<fp16, [1, 32, 64, 32]> var_922 = mul(x = x1_13, y = cos_7)[name = string("op_922")];
            tensor<fp16, [1, 32, 64, 32]> var_923 = mul(x = x2_13, y = sin_7)[name = string("op_923")];
            tensor<fp16, [1, 32, 64, 32]> var_924 = sub(x = var_922, y = var_923)[name = string("op_924")];
            tensor<fp16, [1, 32, 64, 32]> var_925 = mul(x = x2_13, y = cos_7)[name = string("op_925")];
            tensor<fp16, [1, 32, 64, 32]> var_926 = mul(x = x1_13, y = sin_7)[name = string("op_926")];
            tensor<fp16, [1, 32, 64, 32]> var_927 = add(x = var_925, y = var_926)[name = string("op_927")];
            bool rotated_13_interleave_0 = const()[name = string("rotated_13_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 32, 64, 64]> rotated_13 = concat(axis = var_44, interleave = rotated_13_interleave_0, values = (var_924, var_927))[name = string("rotated_13")];
            tensor<int32, [4]> x1_15_begin_0 = const()[name = string("x1_15_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_15_end_0 = const()[name = string("x1_15_end_0"), val = tensor<int32, [4]>([1, 8, 64, 32])];
            tensor<bool, [4]> x1_15_end_mask_0 = const()[name = string("x1_15_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 64, 64]> x_89 = transpose(perm = var_900, x = var_899)[name = string("transpose_32")];
            tensor<fp16, [1, 8, 64, 32]> x1_15 = slice_by_index(begin = x1_15_begin_0, end = x1_15_end_0, end_mask = x1_15_end_mask_0, x = x_89)[name = string("x1_15")];
            tensor<int32, [4]> x2_15_begin_0 = const()[name = string("x2_15_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_15_end_0 = const()[name = string("x2_15_end_0"), val = tensor<int32, [4]>([1, 8, 64, 64])];
            tensor<bool, [4]> x2_15_end_mask_0 = const()[name = string("x2_15_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 64, 32]> x2_15 = slice_by_index(begin = x2_15_begin_0, end = x2_15_end_0, end_mask = x2_15_end_mask_0, x = x_89)[name = string("x2_15")];
            tensor<fp16, [1, 8, 64, 32]> var_943 = mul(x = x1_15, y = cos_7)[name = string("op_943")];
            tensor<fp16, [1, 8, 64, 32]> var_944 = mul(x = x2_15, y = sin_7)[name = string("op_944")];
            tensor<fp16, [1, 8, 64, 32]> var_945 = sub(x = var_943, y = var_944)[name = string("op_945")];
            tensor<fp16, [1, 8, 64, 32]> var_946 = mul(x = x2_15, y = cos_7)[name = string("op_946")];
            tensor<fp16, [1, 8, 64, 32]> var_947 = mul(x = x1_15, y = sin_7)[name = string("op_947")];
            tensor<fp16, [1, 8, 64, 32]> var_948 = add(x = var_946, y = var_947)[name = string("op_948")];
            bool rotated_15_interleave_0 = const()[name = string("rotated_15_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 8, 64, 64]> rotated_15 = concat(axis = var_44, interleave = rotated_15_interleave_0, values = (var_945, var_948))[name = string("rotated_15")];
            tensor<int32, [1]> expand_dims_36 = const()[name = string("expand_dims_36"), val = tensor<int32, [1]>([3])];
            tensor<int32, [1]> expand_dims_37 = const()[name = string("expand_dims_37"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_39 = const()[name = string("expand_dims_39"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_40 = const()[name = string("expand_dims_40"), val = tensor<int32, [1]>([4])];
            int32 concat_56_axis_0 = const()[name = string("concat_56_axis_0"), val = int32(0)];
            bool concat_56_interleave_0 = const()[name = string("concat_56_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_56 = concat(axis = concat_56_axis_0, interleave = concat_56_interleave_0, values = (expand_dims_36, expand_dims_37, current_pos, expand_dims_39))[name = string("concat_56")];
            tensor<int32, [1]> concat_57_values1_0 = const()[name = string("concat_57_values1_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> concat_57_values3_0 = const()[name = string("concat_57_values3_0"), val = tensor<int32, [1]>([0])];
            int32 concat_57_axis_0 = const()[name = string("concat_57_axis_0"), val = int32(0)];
            bool concat_57_interleave_0 = const()[name = string("concat_57_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_57 = concat(axis = concat_57_axis_0, interleave = concat_57_interleave_0, values = (expand_dims_40, concat_57_values1_0, var_366, concat_57_values3_0))[name = string("concat_57")];
            tensor<int32, [4]> model_model_kv_cache_0_internal_tensor_assign_7_stride_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_7_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_7_begin_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_7_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_7_end_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_7_end_mask_0"), val = tensor<bool, [4]>([false, true, false, true])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_7_squeeze_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_7_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<fp16, [32, 8, 2048, 64]> model_model_kv_cache_0_internal_tensor_assign_7_cast_fp16 = slice_update(begin = concat_56, begin_mask = model_model_kv_cache_0_internal_tensor_assign_7_begin_mask_0, end = concat_57, end_mask = model_model_kv_cache_0_internal_tensor_assign_7_end_mask_0, squeeze_mask = model_model_kv_cache_0_internal_tensor_assign_7_squeeze_mask_0, stride = model_model_kv_cache_0_internal_tensor_assign_7_stride_0, update = rotated_15, x = coreml_update_state_21)[name = string("model_model_kv_cache_0_internal_tensor_assign_7_cast_fp16")];
            write_state(data = model_model_kv_cache_0_internal_tensor_assign_7_cast_fp16, input = model_model_kv_cache_0)[name = string("coreml_update_state_22_write_state")];
            tensor<fp16, [32, 8, 2048, 64]> coreml_update_state_22 = read_state(input = model_model_kv_cache_0)[name = string("coreml_update_state_22")];
            tensor<int32, [1]> expand_dims_42 = const()[name = string("expand_dims_42"), val = tensor<int32, [1]>([19])];
            tensor<int32, [1]> expand_dims_43 = const()[name = string("expand_dims_43"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_45 = const()[name = string("expand_dims_45"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_46 = const()[name = string("expand_dims_46"), val = tensor<int32, [1]>([20])];
            int32 concat_60_axis_0 = const()[name = string("concat_60_axis_0"), val = int32(0)];
            bool concat_60_interleave_0 = const()[name = string("concat_60_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_60 = concat(axis = concat_60_axis_0, interleave = concat_60_interleave_0, values = (expand_dims_42, expand_dims_43, current_pos, expand_dims_45))[name = string("concat_60")];
            tensor<int32, [1]> concat_61_values1_0 = const()[name = string("concat_61_values1_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> concat_61_values3_0 = const()[name = string("concat_61_values3_0"), val = tensor<int32, [1]>([0])];
            int32 concat_61_axis_0 = const()[name = string("concat_61_axis_0"), val = int32(0)];
            bool concat_61_interleave_0 = const()[name = string("concat_61_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_61 = concat(axis = concat_61_axis_0, interleave = concat_61_interleave_0, values = (expand_dims_46, concat_61_values1_0, var_366, concat_61_values3_0))[name = string("concat_61")];
            tensor<int32, [4]> model_model_kv_cache_0_internal_tensor_assign_8_stride_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_8_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_8_begin_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_8_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_8_end_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_8_end_mask_0"), val = tensor<bool, [4]>([false, true, false, true])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_8_squeeze_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_8_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<fp16, [1, 8, 64, 64]> value_states_21 = transpose(perm = var_904, x = var_903)[name = string("transpose_31")];
            tensor<fp16, [32, 8, 2048, 64]> model_model_kv_cache_0_internal_tensor_assign_8_cast_fp16 = slice_update(begin = concat_60, begin_mask = model_model_kv_cache_0_internal_tensor_assign_8_begin_mask_0, end = concat_61, end_mask = model_model_kv_cache_0_internal_tensor_assign_8_end_mask_0, squeeze_mask = model_model_kv_cache_0_internal_tensor_assign_8_squeeze_mask_0, stride = model_model_kv_cache_0_internal_tensor_assign_8_stride_0, update = value_states_21, x = coreml_update_state_22)[name = string("model_model_kv_cache_0_internal_tensor_assign_8_cast_fp16")];
            write_state(data = model_model_kv_cache_0_internal_tensor_assign_8_cast_fp16, input = model_model_kv_cache_0)[name = string("coreml_update_state_23_write_state")];
            tensor<fp16, [32, 8, 2048, 64]> coreml_update_state_23 = read_state(input = model_model_kv_cache_0)[name = string("coreml_update_state_23")];
            tensor<int32, [4]> var_971_begin_0 = const()[name = string("op_971_begin_0"), val = tensor<int32, [4]>([3, 0, 0, 0])];
            tensor<int32, [4]> var_971_end_0 = const()[name = string("op_971_end_0"), val = tensor<int32, [4]>([4, 8, 2048, 64])];
            tensor<bool, [4]> var_971_end_mask_0 = const()[name = string("op_971_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<fp16, [1, 8, 2048, 64]> var_971_cast_fp16 = slice_by_index(begin = var_971_begin_0, end = var_971_end_0, end_mask = var_971_end_mask_0, x = coreml_update_state_23)[name = string("op_971_cast_fp16")];
            tensor<int32, [1]> K_layer_cache_7_axes_0 = const()[name = string("K_layer_cache_7_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<fp16, [8, 2048, 64]> K_layer_cache_7_cast_fp16 = squeeze(axes = K_layer_cache_7_axes_0, x = var_971_cast_fp16)[name = string("K_layer_cache_7_cast_fp16")];
            tensor<int32, [4]> var_973_begin_0 = const()[name = string("op_973_begin_0"), val = tensor<int32, [4]>([19, 0, 0, 0])];
            tensor<int32, [4]> var_973_end_0 = const()[name = string("op_973_end_0"), val = tensor<int32, [4]>([20, 8, 2048, 64])];
            tensor<bool, [4]> var_973_end_mask_0 = const()[name = string("op_973_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<fp16, [1, 8, 2048, 64]> var_973_cast_fp16 = slice_by_index(begin = var_973_begin_0, end = var_973_end_0, end_mask = var_973_end_mask_0, x = coreml_update_state_23)[name = string("op_973_cast_fp16")];
            tensor<int32, [1]> V_layer_cache_7_axes_0 = const()[name = string("V_layer_cache_7_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<fp16, [8, 2048, 64]> V_layer_cache_7_cast_fp16 = squeeze(axes = V_layer_cache_7_axes_0, x = var_973_cast_fp16)[name = string("V_layer_cache_7_cast_fp16")];
            tensor<int32, [1]> x_95_axes_0 = const()[name = string("x_95_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [8, 1, 2048, 64]> x_95_cast_fp16 = expand_dims(axes = x_95_axes_0, x = K_layer_cache_7_cast_fp16)[name = string("x_95_cast_fp16")];
            tensor<int32, [4]> var_982 = const()[name = string("op_982"), val = tensor<int32, [4]>([1, 4, 1, 1])];
            tensor<fp16, [8, 4, 2048, 64]> x_97_cast_fp16 = tile(reps = var_982, x = x_95_cast_fp16)[name = string("x_97_cast_fp16")];
            tensor<int32, [4]> var_986 = const()[name = string("op_986"), val = tensor<int32, [4]>([1, -1, 2048, 64])];
            tensor<fp16, [1, 32, 2048, 64]> var_987_cast_fp16 = reshape(shape = var_986, x = x_97_cast_fp16)[name = string("op_987_cast_fp16")];
            tensor<int32, [1]> x_101_axes_0 = const()[name = string("x_101_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [8, 1, 2048, 64]> x_101_cast_fp16 = expand_dims(axes = x_101_axes_0, x = V_layer_cache_7_cast_fp16)[name = string("x_101_cast_fp16")];
            tensor<int32, [4]> var_989 = const()[name = string("op_989"), val = tensor<int32, [4]>([1, 4, 1, 1])];
            tensor<fp16, [8, 4, 2048, 64]> x_103_cast_fp16 = tile(reps = var_989, x = x_101_cast_fp16)[name = string("x_103_cast_fp16")];
            bool var_996_transpose_x_0 = const()[name = string("op_996_transpose_x_0"), val = bool(false)];
            bool var_996_transpose_y_0 = const()[name = string("op_996_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 32, 64, 2048]> var_996_cast_fp16 = matmul(transpose_x = var_996_transpose_x_0, transpose_y = var_996_transpose_y_0, x = rotated_13, y = var_987_cast_fp16)[name = string("op_996_cast_fp16")];
            fp16 var_997_to_fp16 = const()[name = string("op_997_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 32, 64, 2048]> attn_weights_7_cast_fp16 = mul(x = var_996_cast_fp16, y = var_997_to_fp16)[name = string("attn_weights_7_cast_fp16")];
            tensor<fp16, [1, 32, 64, 2048]> x_105_cast_fp16 = add(x = attn_weights_7_cast_fp16, y = causal_mask)[name = string("x_105_cast_fp16")];
            tensor<int32, [1]> reduce_max_3_axes_0 = const()[name = string("reduce_max_3_axes_0"), val = tensor<int32, [1]>([-1])];
            bool reduce_max_3_keep_dims_0 = const()[name = string("reduce_max_3_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 64, 1]> reduce_max_3_cast_fp16 = reduce_max(axes = reduce_max_3_axes_0, keep_dims = reduce_max_3_keep_dims_0, x = x_105_cast_fp16)[name = string("reduce_max_3_cast_fp16")];
            tensor<fp16, [1, 32, 64, 2048]> x_107_cast_fp16 = sub(x = x_105_cast_fp16, y = reduce_max_3_cast_fp16)[name = string("x_107_cast_fp16")];
            tensor<fp16, [1, 32, 64, 2048]> exp_x_7_cast_fp16 = exp(x = x_107_cast_fp16)[name = string("exp_x_7_cast_fp16")];
            tensor<int32, [1]> var_1008_axes_0 = const()[name = string("op_1008_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1008_keep_dims_0 = const()[name = string("op_1008_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 64, 1]> var_1008_cast_fp16 = reduce_sum(axes = var_1008_axes_0, keep_dims = var_1008_keep_dims_0, x = exp_x_7_cast_fp16)[name = string("op_1008_cast_fp16")];
            tensor<fp16, [1, 32, 64, 2048]> var_1009_cast_fp16 = real_div(x = exp_x_7_cast_fp16, y = var_1008_cast_fp16)[name = string("op_1009_cast_fp16")];
            tensor<int32, [3]> concat_66 = const()[name = string("concat_66"), val = tensor<int32, [3]>([32, 64, 2048])];
            tensor<fp16, [32, 64, 2048]> reshape_9_cast_fp16 = reshape(shape = concat_66, x = var_1009_cast_fp16)[name = string("reshape_9_cast_fp16")];
            tensor<int32, [3]> concat_67 = const()[name = string("concat_67"), val = tensor<int32, [3]>([32, 2048, 64])];
            tensor<fp16, [32, 2048, 64]> reshape_10_cast_fp16 = reshape(shape = concat_67, x = x_103_cast_fp16)[name = string("reshape_10_cast_fp16")];
            bool matmul_3_transpose_x_0 = const()[name = string("matmul_3_transpose_x_0"), val = bool(false)];
            bool matmul_3_transpose_y_0 = const()[name = string("matmul_3_transpose_y_0"), val = bool(false)];
            tensor<fp16, [32, 64, 64]> matmul_3_cast_fp16 = matmul(transpose_x = matmul_3_transpose_x_0, transpose_y = matmul_3_transpose_y_0, x = reshape_9_cast_fp16, y = reshape_10_cast_fp16)[name = string("matmul_3_cast_fp16")];
            tensor<int32, [4]> concat_71 = const()[name = string("concat_71"), val = tensor<int32, [4]>([1, 32, 64, 64])];
            tensor<fp16, [1, 32, 64, 64]> reshape_11_cast_fp16 = reshape(shape = concat_71, x = matmul_3_cast_fp16)[name = string("reshape_11_cast_fp16")];
            tensor<int32, [4]> var_1012_perm_0 = const()[name = string("op_1012_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1014 = const()[name = string("op_1014"), val = tensor<int32, [3]>([1, 64, 2048])];
            tensor<fp16, [1, 64, 32, 64]> var_1012_cast_fp16 = transpose(perm = var_1012_perm_0, x = reshape_11_cast_fp16)[name = string("transpose_30")];
            tensor<fp16, [1, 64, 2048]> input_47_cast_fp16 = reshape(shape = var_1014, x = var_1012_cast_fp16)[name = string("input_47_cast_fp16")];
            tensor<fp16, [2048, 2048]> model_model_layers_3_self_attn_o_proj_weight_promoted_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267091008))), lut = tensor<fp16, [256, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269188224))))[name = string("model_model_layers_3_self_attn_o_proj_weight_promoted_to_fp16_palettized")];
            tensor<fp16, [1, 64, 2048]> linear_3_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_3_self_attn_o_proj_weight_promoted_to_fp16_palettized, x = input_47_cast_fp16)[name = string("linear_3_cast_fp16")];
            tensor<fp16, [1, 64, 2048]> hidden_states_29_cast_fp16 = add(x = hidden_states_25_cast_fp16, y = linear_3_cast_fp16)[name = string("hidden_states_29_cast_fp16")];
            tensor<int32, [1]> mean_15_axes_0 = const()[name = string("mean_15_axes_0"), val = tensor<int32, [1]>([-1])];
            bool mean_15_keep_dims_0 = const()[name = string("mean_15_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 64, 1]> mean_15_cast_fp16 = reduce_mean(axes = mean_15_axes_0, keep_dims = mean_15_keep_dims_0, x = hidden_states_29_cast_fp16)[name = string("mean_15_cast_fp16")];
            tensor<fp16, [1, 64, 2048]> input_49_cast_fp16 = sub(x = hidden_states_29_cast_fp16, y = mean_15_cast_fp16)[name = string("input_49_cast_fp16")];
            tensor<int32, [1]> var_1025_axes_0 = const()[name = string("op_1025_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2048]> model_model_layers_3_post_attention_layernorm_weight_to_fp16 = const()[name = string("model_model_layers_3_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269196480)))];
            tensor<fp16, [1, 64, 2048]> var_1025_cast_fp16 = layer_norm(axes = var_1025_axes_0, epsilon = var_46_to_fp16, gamma = model_model_layers_3_post_attention_layernorm_weight_to_fp16, x = input_49_cast_fp16)[name = string("op_1025_cast_fp16")];
            tensor<int32, [3]> var_1032 = const()[name = string("op_1032"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> input_51_axes_0 = const()[name = string("input_51_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 64]> var_1033 = transpose(perm = var_1032, x = var_1025_cast_fp16)[name = string("transpose_29")];
            tensor<fp16, [1, 2048, 1, 64]> input_51 = expand_dims(axes = input_51_axes_0, x = var_1033)[name = string("input_51")];
            string input_53_pad_type_0 = const()[name = string("input_53_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_53_strides_0 = const()[name = string("input_53_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_53_pad_0 = const()[name = string("input_53_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_53_dilations_0 = const()[name = string("input_53_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_53_groups_0 = const()[name = string("input_53_groups_0"), val = int32(1)];
            tensor<fp16, [1, 8192, 1, 64]> input_53 = conv(dilations = input_53_dilations_0, groups = input_53_groups_0, pad = input_53_pad_0, pad_type = input_53_pad_type_0, strides = input_53_strides_0, weight = model_model_layers_3_mlp_gate_proj_weight_palettized, x = input_51)[name = string("input_53")];
            string up_states_7_pad_type_0 = const()[name = string("up_states_7_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> up_states_7_strides_0 = const()[name = string("up_states_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> up_states_7_pad_0 = const()[name = string("up_states_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> up_states_7_dilations_0 = const()[name = string("up_states_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 up_states_7_groups_0 = const()[name = string("up_states_7_groups_0"), val = int32(1)];
            tensor<fp16, [1, 8192, 1, 64]> up_states_7 = conv(dilations = up_states_7_dilations_0, groups = up_states_7_groups_0, pad = up_states_7_pad_0, pad_type = up_states_7_pad_type_0, strides = up_states_7_strides_0, weight = model_model_layers_3_mlp_up_proj_weight_palettized, x = input_51)[name = string("up_states_7")];
            tensor<fp16, [1, 8192, 1, 64]> gate_states_7 = silu(x = input_53)[name = string("gate_states_7")];
            tensor<fp16, [1, 8192, 1, 64]> input_55 = mul(x = gate_states_7, y = up_states_7)[name = string("input_55")];
            string hidden_states_31_pad_type_0 = const()[name = string("hidden_states_31_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_31_strides_0 = const()[name = string("hidden_states_31_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_31_pad_0 = const()[name = string("hidden_states_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_31_dilations_0 = const()[name = string("hidden_states_31_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_31_groups_0 = const()[name = string("hidden_states_31_groups_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 1, 64]> hidden_states_31 = conv(dilations = hidden_states_31_dilations_0, groups = hidden_states_31_groups_0, pad = hidden_states_31_pad_0, pad_type = hidden_states_31_pad_type_0, strides = hidden_states_31_strides_0, weight = model_model_layers_3_mlp_down_proj_weight_palettized, x = input_55)[name = string("hidden_states_31")];
            tensor<int32, [1]> var_1055_axes_0 = const()[name = string("op_1055_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 64]> var_1055 = squeeze(axes = var_1055_axes_0, x = hidden_states_31)[name = string("op_1055")];
            tensor<int32, [3]> var_1056 = const()[name = string("op_1056"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 64, 2048]> var_1057 = transpose(perm = var_1056, x = var_1055)[name = string("transpose_28")];
            tensor<fp16, [1, 64, 2048]> hidden_states_33_cast_fp16 = add(x = hidden_states_29_cast_fp16, y = var_1057)[name = string("hidden_states_33_cast_fp16")];
            tensor<int32, [1]> mean_17_axes_0 = const()[name = string("mean_17_axes_0"), val = tensor<int32, [1]>([-1])];
            bool mean_17_keep_dims_0 = const()[name = string("mean_17_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 64, 1]> mean_17_cast_fp16 = reduce_mean(axes = mean_17_axes_0, keep_dims = mean_17_keep_dims_0, x = hidden_states_33_cast_fp16)[name = string("mean_17_cast_fp16")];
            tensor<fp16, [1, 64, 2048]> input_57_cast_fp16 = sub(x = hidden_states_33_cast_fp16, y = mean_17_cast_fp16)[name = string("input_57_cast_fp16")];
            tensor<int32, [1]> var_1065_axes_0 = const()[name = string("op_1065_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2048]> model_model_layers_4_input_layernorm_weight_to_fp16 = const()[name = string("model_model_layers_4_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269200640)))];
            tensor<fp16, [1, 64, 2048]> var_1065_cast_fp16 = layer_norm(axes = var_1065_axes_0, epsilon = var_46_to_fp16, gamma = model_model_layers_4_input_layernorm_weight_to_fp16, x = input_57_cast_fp16)[name = string("op_1065_cast_fp16")];
            tensor<int32, [3]> var_1069 = const()[name = string("op_1069"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1071_axes_0 = const()[name = string("op_1071_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 64]> var_1070 = transpose(perm = var_1069, x = var_1065_cast_fp16)[name = string("transpose_27")];
            tensor<fp16, [1, 2048, 1, 64]> var_1071 = expand_dims(axes = var_1071_axes_0, x = var_1070)[name = string("op_1071")];
            string query_states_17_pad_type_0 = const()[name = string("query_states_17_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_states_17_strides_0 = const()[name = string("query_states_17_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_states_17_pad_0 = const()[name = string("query_states_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_states_17_dilations_0 = const()[name = string("query_states_17_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_states_17_groups_0 = const()[name = string("query_states_17_groups_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 1, 64]> query_states_17 = conv(dilations = query_states_17_dilations_0, groups = query_states_17_groups_0, pad = query_states_17_pad_0, pad_type = query_states_17_pad_type_0, strides = query_states_17_strides_0, weight = model_model_layers_4_self_attn_q_proj_weight_palettized, x = var_1071)[name = string("query_states_17")];
            string key_states_25_pad_type_0 = const()[name = string("key_states_25_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_states_25_strides_0 = const()[name = string("key_states_25_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_states_25_pad_0 = const()[name = string("key_states_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_states_25_dilations_0 = const()[name = string("key_states_25_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_states_25_groups_0 = const()[name = string("key_states_25_groups_0"), val = int32(1)];
            tensor<fp16, [1, 512, 1, 64]> key_states_25 = conv(dilations = key_states_25_dilations_0, groups = key_states_25_groups_0, pad = key_states_25_pad_0, pad_type = key_states_25_pad_type_0, strides = key_states_25_strides_0, weight = model_model_layers_4_self_attn_k_proj_weight_palettized, x = var_1071)[name = string("key_states_25")];
            string value_states_25_pad_type_0 = const()[name = string("value_states_25_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_states_25_strides_0 = const()[name = string("value_states_25_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_states_25_pad_0 = const()[name = string("value_states_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_states_25_dilations_0 = const()[name = string("value_states_25_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_states_25_groups_0 = const()[name = string("value_states_25_groups_0"), val = int32(1)];
            tensor<fp16, [1, 512, 1, 64]> value_states_25 = conv(dilations = value_states_25_dilations_0, groups = value_states_25_groups_0, pad = value_states_25_pad_0, pad_type = value_states_25_pad_type_0, strides = value_states_25_strides_0, weight = model_model_layers_4_self_attn_v_proj_weight_palettized, x = var_1071)[name = string("value_states_25")];
            tensor<int32, [4]> var_1091 = const()[name = string("op_1091"), val = tensor<int32, [4]>([1, 32, 64, 64])];
            tensor<fp16, [1, 32, 64, 64]> var_1092 = reshape(shape = var_1091, x = query_states_17)[name = string("op_1092")];
            tensor<int32, [4]> var_1093 = const()[name = string("op_1093"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_1095 = const()[name = string("op_1095"), val = tensor<int32, [4]>([1, 8, 64, 64])];
            tensor<fp16, [1, 8, 64, 64]> var_1096 = reshape(shape = var_1095, x = key_states_25)[name = string("op_1096")];
            tensor<int32, [4]> var_1097 = const()[name = string("op_1097"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_1099 = const()[name = string("op_1099"), val = tensor<int32, [4]>([1, 8, 64, 64])];
            tensor<fp16, [1, 8, 64, 64]> var_1100 = reshape(shape = var_1099, x = value_states_25)[name = string("op_1100")];
            tensor<int32, [4]> var_1101 = const()[name = string("op_1101"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> x1_17_begin_0 = const()[name = string("x1_17_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_17_end_0 = const()[name = string("x1_17_end_0"), val = tensor<int32, [4]>([1, 32, 64, 32])];
            tensor<bool, [4]> x1_17_end_mask_0 = const()[name = string("x1_17_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 64, 64]> x_113 = transpose(perm = var_1093, x = var_1092)[name = string("transpose_26")];
            tensor<fp16, [1, 32, 64, 32]> x1_17 = slice_by_index(begin = x1_17_begin_0, end = x1_17_end_0, end_mask = x1_17_end_mask_0, x = x_113)[name = string("x1_17")];
            tensor<int32, [4]> x2_17_begin_0 = const()[name = string("x2_17_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_17_end_0 = const()[name = string("x2_17_end_0"), val = tensor<int32, [4]>([1, 32, 64, 64])];
            tensor<bool, [4]> x2_17_end_mask_0 = const()[name = string("x2_17_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 64, 32]> x2_17 = slice_by_index(begin = x2_17_begin_0, end = x2_17_end_0, end_mask = x2_17_end_mask_0, x = x_113)[name = string("x2_17")];
            tensor<fp16, [1, 32, 64, 32]> var_1119 = mul(x = x1_17, y = cos_7)[name = string("op_1119")];
            tensor<fp16, [1, 32, 64, 32]> var_1120 = mul(x = x2_17, y = sin_7)[name = string("op_1120")];
            tensor<fp16, [1, 32, 64, 32]> var_1121 = sub(x = var_1119, y = var_1120)[name = string("op_1121")];
            tensor<fp16, [1, 32, 64, 32]> var_1122 = mul(x = x2_17, y = cos_7)[name = string("op_1122")];
            tensor<fp16, [1, 32, 64, 32]> var_1123 = mul(x = x1_17, y = sin_7)[name = string("op_1123")];
            tensor<fp16, [1, 32, 64, 32]> var_1124 = add(x = var_1122, y = var_1123)[name = string("op_1124")];
            bool rotated_17_interleave_0 = const()[name = string("rotated_17_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 32, 64, 64]> rotated_17 = concat(axis = var_44, interleave = rotated_17_interleave_0, values = (var_1121, var_1124))[name = string("rotated_17")];
            tensor<int32, [4]> x1_19_begin_0 = const()[name = string("x1_19_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_19_end_0 = const()[name = string("x1_19_end_0"), val = tensor<int32, [4]>([1, 8, 64, 32])];
            tensor<bool, [4]> x1_19_end_mask_0 = const()[name = string("x1_19_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 64, 64]> x_117 = transpose(perm = var_1097, x = var_1096)[name = string("transpose_25")];
            tensor<fp16, [1, 8, 64, 32]> x1_19 = slice_by_index(begin = x1_19_begin_0, end = x1_19_end_0, end_mask = x1_19_end_mask_0, x = x_117)[name = string("x1_19")];
            tensor<int32, [4]> x2_19_begin_0 = const()[name = string("x2_19_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_19_end_0 = const()[name = string("x2_19_end_0"), val = tensor<int32, [4]>([1, 8, 64, 64])];
            tensor<bool, [4]> x2_19_end_mask_0 = const()[name = string("x2_19_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 64, 32]> x2_19 = slice_by_index(begin = x2_19_begin_0, end = x2_19_end_0, end_mask = x2_19_end_mask_0, x = x_117)[name = string("x2_19")];
            tensor<fp16, [1, 8, 64, 32]> var_1140 = mul(x = x1_19, y = cos_7)[name = string("op_1140")];
            tensor<fp16, [1, 8, 64, 32]> var_1141 = mul(x = x2_19, y = sin_7)[name = string("op_1141")];
            tensor<fp16, [1, 8, 64, 32]> var_1142 = sub(x = var_1140, y = var_1141)[name = string("op_1142")];
            tensor<fp16, [1, 8, 64, 32]> var_1143 = mul(x = x2_19, y = cos_7)[name = string("op_1143")];
            tensor<fp16, [1, 8, 64, 32]> var_1144 = mul(x = x1_19, y = sin_7)[name = string("op_1144")];
            tensor<fp16, [1, 8, 64, 32]> var_1145 = add(x = var_1143, y = var_1144)[name = string("op_1145")];
            bool rotated_19_interleave_0 = const()[name = string("rotated_19_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 8, 64, 64]> rotated_19 = concat(axis = var_44, interleave = rotated_19_interleave_0, values = (var_1142, var_1145))[name = string("rotated_19")];
            tensor<int32, [1]> expand_dims_48 = const()[name = string("expand_dims_48"), val = tensor<int32, [1]>([4])];
            tensor<int32, [1]> expand_dims_49 = const()[name = string("expand_dims_49"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_51 = const()[name = string("expand_dims_51"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_52 = const()[name = string("expand_dims_52"), val = tensor<int32, [1]>([5])];
            int32 concat_74_axis_0 = const()[name = string("concat_74_axis_0"), val = int32(0)];
            bool concat_74_interleave_0 = const()[name = string("concat_74_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_74 = concat(axis = concat_74_axis_0, interleave = concat_74_interleave_0, values = (expand_dims_48, expand_dims_49, current_pos, expand_dims_51))[name = string("concat_74")];
            tensor<int32, [1]> concat_75_values1_0 = const()[name = string("concat_75_values1_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> concat_75_values3_0 = const()[name = string("concat_75_values3_0"), val = tensor<int32, [1]>([0])];
            int32 concat_75_axis_0 = const()[name = string("concat_75_axis_0"), val = int32(0)];
            bool concat_75_interleave_0 = const()[name = string("concat_75_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_75 = concat(axis = concat_75_axis_0, interleave = concat_75_interleave_0, values = (expand_dims_52, concat_75_values1_0, var_366, concat_75_values3_0))[name = string("concat_75")];
            tensor<int32, [4]> model_model_kv_cache_0_internal_tensor_assign_9_stride_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_9_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_9_begin_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_9_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_9_end_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_9_end_mask_0"), val = tensor<bool, [4]>([false, true, false, true])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_9_squeeze_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_9_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<fp16, [32, 8, 2048, 64]> model_model_kv_cache_0_internal_tensor_assign_9_cast_fp16 = slice_update(begin = concat_74, begin_mask = model_model_kv_cache_0_internal_tensor_assign_9_begin_mask_0, end = concat_75, end_mask = model_model_kv_cache_0_internal_tensor_assign_9_end_mask_0, squeeze_mask = model_model_kv_cache_0_internal_tensor_assign_9_squeeze_mask_0, stride = model_model_kv_cache_0_internal_tensor_assign_9_stride_0, update = rotated_19, x = coreml_update_state_23)[name = string("model_model_kv_cache_0_internal_tensor_assign_9_cast_fp16")];
            write_state(data = model_model_kv_cache_0_internal_tensor_assign_9_cast_fp16, input = model_model_kv_cache_0)[name = string("coreml_update_state_24_write_state")];
            tensor<fp16, [32, 8, 2048, 64]> coreml_update_state_24 = read_state(input = model_model_kv_cache_0)[name = string("coreml_update_state_24")];
            tensor<int32, [1]> expand_dims_54 = const()[name = string("expand_dims_54"), val = tensor<int32, [1]>([20])];
            tensor<int32, [1]> expand_dims_55 = const()[name = string("expand_dims_55"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_57 = const()[name = string("expand_dims_57"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_58 = const()[name = string("expand_dims_58"), val = tensor<int32, [1]>([21])];
            int32 concat_78_axis_0 = const()[name = string("concat_78_axis_0"), val = int32(0)];
            bool concat_78_interleave_0 = const()[name = string("concat_78_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_78 = concat(axis = concat_78_axis_0, interleave = concat_78_interleave_0, values = (expand_dims_54, expand_dims_55, current_pos, expand_dims_57))[name = string("concat_78")];
            tensor<int32, [1]> concat_79_values1_0 = const()[name = string("concat_79_values1_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> concat_79_values3_0 = const()[name = string("concat_79_values3_0"), val = tensor<int32, [1]>([0])];
            int32 concat_79_axis_0 = const()[name = string("concat_79_axis_0"), val = int32(0)];
            bool concat_79_interleave_0 = const()[name = string("concat_79_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_79 = concat(axis = concat_79_axis_0, interleave = concat_79_interleave_0, values = (expand_dims_58, concat_79_values1_0, var_366, concat_79_values3_0))[name = string("concat_79")];
            tensor<int32, [4]> model_model_kv_cache_0_internal_tensor_assign_10_stride_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_10_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_10_begin_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_10_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_10_end_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_10_end_mask_0"), val = tensor<bool, [4]>([false, true, false, true])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_10_squeeze_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_10_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<fp16, [1, 8, 64, 64]> value_states_27 = transpose(perm = var_1101, x = var_1100)[name = string("transpose_24")];
            tensor<fp16, [32, 8, 2048, 64]> model_model_kv_cache_0_internal_tensor_assign_10_cast_fp16 = slice_update(begin = concat_78, begin_mask = model_model_kv_cache_0_internal_tensor_assign_10_begin_mask_0, end = concat_79, end_mask = model_model_kv_cache_0_internal_tensor_assign_10_end_mask_0, squeeze_mask = model_model_kv_cache_0_internal_tensor_assign_10_squeeze_mask_0, stride = model_model_kv_cache_0_internal_tensor_assign_10_stride_0, update = value_states_27, x = coreml_update_state_24)[name = string("model_model_kv_cache_0_internal_tensor_assign_10_cast_fp16")];
            write_state(data = model_model_kv_cache_0_internal_tensor_assign_10_cast_fp16, input = model_model_kv_cache_0)[name = string("coreml_update_state_25_write_state")];
            tensor<fp16, [32, 8, 2048, 64]> coreml_update_state_25 = read_state(input = model_model_kv_cache_0)[name = string("coreml_update_state_25")];
            tensor<int32, [4]> var_1168_begin_0 = const()[name = string("op_1168_begin_0"), val = tensor<int32, [4]>([4, 0, 0, 0])];
            tensor<int32, [4]> var_1168_end_0 = const()[name = string("op_1168_end_0"), val = tensor<int32, [4]>([5, 8, 2048, 64])];
            tensor<bool, [4]> var_1168_end_mask_0 = const()[name = string("op_1168_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<fp16, [1, 8, 2048, 64]> var_1168_cast_fp16 = slice_by_index(begin = var_1168_begin_0, end = var_1168_end_0, end_mask = var_1168_end_mask_0, x = coreml_update_state_25)[name = string("op_1168_cast_fp16")];
            tensor<int32, [1]> K_layer_cache_9_axes_0 = const()[name = string("K_layer_cache_9_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<fp16, [8, 2048, 64]> K_layer_cache_9_cast_fp16 = squeeze(axes = K_layer_cache_9_axes_0, x = var_1168_cast_fp16)[name = string("K_layer_cache_9_cast_fp16")];
            tensor<int32, [4]> var_1170_begin_0 = const()[name = string("op_1170_begin_0"), val = tensor<int32, [4]>([20, 0, 0, 0])];
            tensor<int32, [4]> var_1170_end_0 = const()[name = string("op_1170_end_0"), val = tensor<int32, [4]>([21, 8, 2048, 64])];
            tensor<bool, [4]> var_1170_end_mask_0 = const()[name = string("op_1170_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<fp16, [1, 8, 2048, 64]> var_1170_cast_fp16 = slice_by_index(begin = var_1170_begin_0, end = var_1170_end_0, end_mask = var_1170_end_mask_0, x = coreml_update_state_25)[name = string("op_1170_cast_fp16")];
            tensor<int32, [1]> V_layer_cache_9_axes_0 = const()[name = string("V_layer_cache_9_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<fp16, [8, 2048, 64]> V_layer_cache_9_cast_fp16 = squeeze(axes = V_layer_cache_9_axes_0, x = var_1170_cast_fp16)[name = string("V_layer_cache_9_cast_fp16")];
            tensor<int32, [1]> x_123_axes_0 = const()[name = string("x_123_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [8, 1, 2048, 64]> x_123_cast_fp16 = expand_dims(axes = x_123_axes_0, x = K_layer_cache_9_cast_fp16)[name = string("x_123_cast_fp16")];
            tensor<int32, [4]> var_1179 = const()[name = string("op_1179"), val = tensor<int32, [4]>([1, 4, 1, 1])];
            tensor<fp16, [8, 4, 2048, 64]> x_125_cast_fp16 = tile(reps = var_1179, x = x_123_cast_fp16)[name = string("x_125_cast_fp16")];
            tensor<int32, [4]> var_1183 = const()[name = string("op_1183"), val = tensor<int32, [4]>([1, -1, 2048, 64])];
            tensor<fp16, [1, 32, 2048, 64]> var_1184_cast_fp16 = reshape(shape = var_1183, x = x_125_cast_fp16)[name = string("op_1184_cast_fp16")];
            tensor<int32, [1]> x_129_axes_0 = const()[name = string("x_129_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [8, 1, 2048, 64]> x_129_cast_fp16 = expand_dims(axes = x_129_axes_0, x = V_layer_cache_9_cast_fp16)[name = string("x_129_cast_fp16")];
            tensor<int32, [4]> var_1186 = const()[name = string("op_1186"), val = tensor<int32, [4]>([1, 4, 1, 1])];
            tensor<fp16, [8, 4, 2048, 64]> x_131_cast_fp16 = tile(reps = var_1186, x = x_129_cast_fp16)[name = string("x_131_cast_fp16")];
            bool var_1193_transpose_x_0 = const()[name = string("op_1193_transpose_x_0"), val = bool(false)];
            bool var_1193_transpose_y_0 = const()[name = string("op_1193_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 32, 64, 2048]> var_1193_cast_fp16 = matmul(transpose_x = var_1193_transpose_x_0, transpose_y = var_1193_transpose_y_0, x = rotated_17, y = var_1184_cast_fp16)[name = string("op_1193_cast_fp16")];
            fp16 var_1194_to_fp16 = const()[name = string("op_1194_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 32, 64, 2048]> attn_weights_9_cast_fp16 = mul(x = var_1193_cast_fp16, y = var_1194_to_fp16)[name = string("attn_weights_9_cast_fp16")];
            tensor<fp16, [1, 32, 64, 2048]> x_133_cast_fp16 = add(x = attn_weights_9_cast_fp16, y = causal_mask)[name = string("x_133_cast_fp16")];
            tensor<int32, [1]> reduce_max_4_axes_0 = const()[name = string("reduce_max_4_axes_0"), val = tensor<int32, [1]>([-1])];
            bool reduce_max_4_keep_dims_0 = const()[name = string("reduce_max_4_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 64, 1]> reduce_max_4_cast_fp16 = reduce_max(axes = reduce_max_4_axes_0, keep_dims = reduce_max_4_keep_dims_0, x = x_133_cast_fp16)[name = string("reduce_max_4_cast_fp16")];
            tensor<fp16, [1, 32, 64, 2048]> x_135_cast_fp16 = sub(x = x_133_cast_fp16, y = reduce_max_4_cast_fp16)[name = string("x_135_cast_fp16")];
            tensor<fp16, [1, 32, 64, 2048]> exp_x_9_cast_fp16 = exp(x = x_135_cast_fp16)[name = string("exp_x_9_cast_fp16")];
            tensor<int32, [1]> var_1205_axes_0 = const()[name = string("op_1205_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1205_keep_dims_0 = const()[name = string("op_1205_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 64, 1]> var_1205_cast_fp16 = reduce_sum(axes = var_1205_axes_0, keep_dims = var_1205_keep_dims_0, x = exp_x_9_cast_fp16)[name = string("op_1205_cast_fp16")];
            tensor<fp16, [1, 32, 64, 2048]> var_1206_cast_fp16 = real_div(x = exp_x_9_cast_fp16, y = var_1205_cast_fp16)[name = string("op_1206_cast_fp16")];
            tensor<int32, [3]> concat_84 = const()[name = string("concat_84"), val = tensor<int32, [3]>([32, 64, 2048])];
            tensor<fp16, [32, 64, 2048]> reshape_12_cast_fp16 = reshape(shape = concat_84, x = var_1206_cast_fp16)[name = string("reshape_12_cast_fp16")];
            tensor<int32, [3]> concat_85 = const()[name = string("concat_85"), val = tensor<int32, [3]>([32, 2048, 64])];
            tensor<fp16, [32, 2048, 64]> reshape_13_cast_fp16 = reshape(shape = concat_85, x = x_131_cast_fp16)[name = string("reshape_13_cast_fp16")];
            bool matmul_4_transpose_x_0 = const()[name = string("matmul_4_transpose_x_0"), val = bool(false)];
            bool matmul_4_transpose_y_0 = const()[name = string("matmul_4_transpose_y_0"), val = bool(false)];
            tensor<fp16, [32, 64, 64]> matmul_4_cast_fp16 = matmul(transpose_x = matmul_4_transpose_x_0, transpose_y = matmul_4_transpose_y_0, x = reshape_12_cast_fp16, y = reshape_13_cast_fp16)[name = string("matmul_4_cast_fp16")];
            tensor<int32, [4]> concat_89 = const()[name = string("concat_89"), val = tensor<int32, [4]>([1, 32, 64, 64])];
            tensor<fp16, [1, 32, 64, 64]> reshape_14_cast_fp16 = reshape(shape = concat_89, x = matmul_4_cast_fp16)[name = string("reshape_14_cast_fp16")];
            tensor<int32, [4]> var_1209_perm_0 = const()[name = string("op_1209_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1211 = const()[name = string("op_1211"), val = tensor<int32, [3]>([1, 64, 2048])];
            tensor<fp16, [1, 64, 32, 64]> var_1209_cast_fp16 = transpose(perm = var_1209_perm_0, x = reshape_14_cast_fp16)[name = string("transpose_23")];
            tensor<fp16, [1, 64, 2048]> input_61_cast_fp16 = reshape(shape = var_1211, x = var_1209_cast_fp16)[name = string("input_61_cast_fp16")];
            tensor<fp16, [2048, 2048]> model_model_layers_4_self_attn_o_proj_weight_promoted_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269204800))), lut = tensor<fp16, [256, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271302016))))[name = string("model_model_layers_4_self_attn_o_proj_weight_promoted_to_fp16_palettized")];
            tensor<fp16, [1, 64, 2048]> linear_4_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_4_self_attn_o_proj_weight_promoted_to_fp16_palettized, x = input_61_cast_fp16)[name = string("linear_4_cast_fp16")];
            tensor<fp16, [1, 64, 2048]> hidden_states_37_cast_fp16 = add(x = hidden_states_33_cast_fp16, y = linear_4_cast_fp16)[name = string("hidden_states_37_cast_fp16")];
            tensor<int32, [1]> mean_19_axes_0 = const()[name = string("mean_19_axes_0"), val = tensor<int32, [1]>([-1])];
            bool mean_19_keep_dims_0 = const()[name = string("mean_19_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 64, 1]> mean_19_cast_fp16 = reduce_mean(axes = mean_19_axes_0, keep_dims = mean_19_keep_dims_0, x = hidden_states_37_cast_fp16)[name = string("mean_19_cast_fp16")];
            tensor<fp16, [1, 64, 2048]> input_63_cast_fp16 = sub(x = hidden_states_37_cast_fp16, y = mean_19_cast_fp16)[name = string("input_63_cast_fp16")];
            tensor<int32, [1]> var_1222_axes_0 = const()[name = string("op_1222_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2048]> model_model_layers_4_post_attention_layernorm_weight_to_fp16 = const()[name = string("model_model_layers_4_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271310272)))];
            tensor<fp16, [1, 64, 2048]> var_1222_cast_fp16 = layer_norm(axes = var_1222_axes_0, epsilon = var_46_to_fp16, gamma = model_model_layers_4_post_attention_layernorm_weight_to_fp16, x = input_63_cast_fp16)[name = string("op_1222_cast_fp16")];
            tensor<int32, [3]> var_1229 = const()[name = string("op_1229"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> input_65_axes_0 = const()[name = string("input_65_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 64]> var_1230 = transpose(perm = var_1229, x = var_1222_cast_fp16)[name = string("transpose_22")];
            tensor<fp16, [1, 2048, 1, 64]> input_65 = expand_dims(axes = input_65_axes_0, x = var_1230)[name = string("input_65")];
            string input_67_pad_type_0 = const()[name = string("input_67_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_67_strides_0 = const()[name = string("input_67_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_67_pad_0 = const()[name = string("input_67_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_67_dilations_0 = const()[name = string("input_67_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_67_groups_0 = const()[name = string("input_67_groups_0"), val = int32(1)];
            tensor<fp16, [1, 8192, 1, 64]> input_67 = conv(dilations = input_67_dilations_0, groups = input_67_groups_0, pad = input_67_pad_0, pad_type = input_67_pad_type_0, strides = input_67_strides_0, weight = model_model_layers_4_mlp_gate_proj_weight_palettized, x = input_65)[name = string("input_67")];
            string up_states_9_pad_type_0 = const()[name = string("up_states_9_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> up_states_9_strides_0 = const()[name = string("up_states_9_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> up_states_9_pad_0 = const()[name = string("up_states_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> up_states_9_dilations_0 = const()[name = string("up_states_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 up_states_9_groups_0 = const()[name = string("up_states_9_groups_0"), val = int32(1)];
            tensor<fp16, [1, 8192, 1, 64]> up_states_9 = conv(dilations = up_states_9_dilations_0, groups = up_states_9_groups_0, pad = up_states_9_pad_0, pad_type = up_states_9_pad_type_0, strides = up_states_9_strides_0, weight = model_model_layers_4_mlp_up_proj_weight_palettized, x = input_65)[name = string("up_states_9")];
            tensor<fp16, [1, 8192, 1, 64]> gate_states_9 = silu(x = input_67)[name = string("gate_states_9")];
            tensor<fp16, [1, 8192, 1, 64]> input_69 = mul(x = gate_states_9, y = up_states_9)[name = string("input_69")];
            string hidden_states_39_pad_type_0 = const()[name = string("hidden_states_39_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_39_strides_0 = const()[name = string("hidden_states_39_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_39_pad_0 = const()[name = string("hidden_states_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_39_dilations_0 = const()[name = string("hidden_states_39_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_39_groups_0 = const()[name = string("hidden_states_39_groups_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 1, 64]> hidden_states_39 = conv(dilations = hidden_states_39_dilations_0, groups = hidden_states_39_groups_0, pad = hidden_states_39_pad_0, pad_type = hidden_states_39_pad_type_0, strides = hidden_states_39_strides_0, weight = model_model_layers_4_mlp_down_proj_weight_palettized, x = input_69)[name = string("hidden_states_39")];
            tensor<int32, [1]> var_1252_axes_0 = const()[name = string("op_1252_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 64]> var_1252 = squeeze(axes = var_1252_axes_0, x = hidden_states_39)[name = string("op_1252")];
            tensor<int32, [3]> var_1253 = const()[name = string("op_1253"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 64, 2048]> var_1254 = transpose(perm = var_1253, x = var_1252)[name = string("transpose_21")];
            tensor<fp16, [1, 64, 2048]> hidden_states_41_cast_fp16 = add(x = hidden_states_37_cast_fp16, y = var_1254)[name = string("hidden_states_41_cast_fp16")];
            tensor<int32, [1]> mean_21_axes_0 = const()[name = string("mean_21_axes_0"), val = tensor<int32, [1]>([-1])];
            bool mean_21_keep_dims_0 = const()[name = string("mean_21_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 64, 1]> mean_21_cast_fp16 = reduce_mean(axes = mean_21_axes_0, keep_dims = mean_21_keep_dims_0, x = hidden_states_41_cast_fp16)[name = string("mean_21_cast_fp16")];
            tensor<fp16, [1, 64, 2048]> input_71_cast_fp16 = sub(x = hidden_states_41_cast_fp16, y = mean_21_cast_fp16)[name = string("input_71_cast_fp16")];
            tensor<int32, [1]> var_1262_axes_0 = const()[name = string("op_1262_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2048]> model_model_layers_5_input_layernorm_weight_to_fp16 = const()[name = string("model_model_layers_5_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271314432)))];
            tensor<fp16, [1, 64, 2048]> var_1262_cast_fp16 = layer_norm(axes = var_1262_axes_0, epsilon = var_46_to_fp16, gamma = model_model_layers_5_input_layernorm_weight_to_fp16, x = input_71_cast_fp16)[name = string("op_1262_cast_fp16")];
            tensor<int32, [3]> var_1266 = const()[name = string("op_1266"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1268_axes_0 = const()[name = string("op_1268_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 64]> var_1267 = transpose(perm = var_1266, x = var_1262_cast_fp16)[name = string("transpose_20")];
            tensor<fp16, [1, 2048, 1, 64]> var_1268 = expand_dims(axes = var_1268_axes_0, x = var_1267)[name = string("op_1268")];
            string query_states_21_pad_type_0 = const()[name = string("query_states_21_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_states_21_strides_0 = const()[name = string("query_states_21_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_states_21_pad_0 = const()[name = string("query_states_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_states_21_dilations_0 = const()[name = string("query_states_21_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_states_21_groups_0 = const()[name = string("query_states_21_groups_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 1, 64]> query_states_21 = conv(dilations = query_states_21_dilations_0, groups = query_states_21_groups_0, pad = query_states_21_pad_0, pad_type = query_states_21_pad_type_0, strides = query_states_21_strides_0, weight = model_model_layers_5_self_attn_q_proj_weight_palettized, x = var_1268)[name = string("query_states_21")];
            string key_states_31_pad_type_0 = const()[name = string("key_states_31_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_states_31_strides_0 = const()[name = string("key_states_31_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_states_31_pad_0 = const()[name = string("key_states_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_states_31_dilations_0 = const()[name = string("key_states_31_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_states_31_groups_0 = const()[name = string("key_states_31_groups_0"), val = int32(1)];
            tensor<fp16, [1, 512, 1, 64]> key_states_31 = conv(dilations = key_states_31_dilations_0, groups = key_states_31_groups_0, pad = key_states_31_pad_0, pad_type = key_states_31_pad_type_0, strides = key_states_31_strides_0, weight = model_model_layers_5_self_attn_k_proj_weight_palettized, x = var_1268)[name = string("key_states_31")];
            string value_states_31_pad_type_0 = const()[name = string("value_states_31_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_states_31_strides_0 = const()[name = string("value_states_31_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_states_31_pad_0 = const()[name = string("value_states_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_states_31_dilations_0 = const()[name = string("value_states_31_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_states_31_groups_0 = const()[name = string("value_states_31_groups_0"), val = int32(1)];
            tensor<fp16, [1, 512, 1, 64]> value_states_31 = conv(dilations = value_states_31_dilations_0, groups = value_states_31_groups_0, pad = value_states_31_pad_0, pad_type = value_states_31_pad_type_0, strides = value_states_31_strides_0, weight = model_model_layers_5_self_attn_v_proj_weight_palettized, x = var_1268)[name = string("value_states_31")];
            tensor<int32, [4]> var_1288 = const()[name = string("op_1288"), val = tensor<int32, [4]>([1, 32, 64, 64])];
            tensor<fp16, [1, 32, 64, 64]> var_1289 = reshape(shape = var_1288, x = query_states_21)[name = string("op_1289")];
            tensor<int32, [4]> var_1290 = const()[name = string("op_1290"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_1292 = const()[name = string("op_1292"), val = tensor<int32, [4]>([1, 8, 64, 64])];
            tensor<fp16, [1, 8, 64, 64]> var_1293 = reshape(shape = var_1292, x = key_states_31)[name = string("op_1293")];
            tensor<int32, [4]> var_1294 = const()[name = string("op_1294"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_1296 = const()[name = string("op_1296"), val = tensor<int32, [4]>([1, 8, 64, 64])];
            tensor<fp16, [1, 8, 64, 64]> var_1297 = reshape(shape = var_1296, x = value_states_31)[name = string("op_1297")];
            tensor<int32, [4]> var_1298 = const()[name = string("op_1298"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> x1_21_begin_0 = const()[name = string("x1_21_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_21_end_0 = const()[name = string("x1_21_end_0"), val = tensor<int32, [4]>([1, 32, 64, 32])];
            tensor<bool, [4]> x1_21_end_mask_0 = const()[name = string("x1_21_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 64, 64]> x_141 = transpose(perm = var_1290, x = var_1289)[name = string("transpose_19")];
            tensor<fp16, [1, 32, 64, 32]> x1_21 = slice_by_index(begin = x1_21_begin_0, end = x1_21_end_0, end_mask = x1_21_end_mask_0, x = x_141)[name = string("x1_21")];
            tensor<int32, [4]> x2_21_begin_0 = const()[name = string("x2_21_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_21_end_0 = const()[name = string("x2_21_end_0"), val = tensor<int32, [4]>([1, 32, 64, 64])];
            tensor<bool, [4]> x2_21_end_mask_0 = const()[name = string("x2_21_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 64, 32]> x2_21 = slice_by_index(begin = x2_21_begin_0, end = x2_21_end_0, end_mask = x2_21_end_mask_0, x = x_141)[name = string("x2_21")];
            tensor<fp16, [1, 32, 64, 32]> var_1316 = mul(x = x1_21, y = cos_7)[name = string("op_1316")];
            tensor<fp16, [1, 32, 64, 32]> var_1317 = mul(x = x2_21, y = sin_7)[name = string("op_1317")];
            tensor<fp16, [1, 32, 64, 32]> var_1318 = sub(x = var_1316, y = var_1317)[name = string("op_1318")];
            tensor<fp16, [1, 32, 64, 32]> var_1319 = mul(x = x2_21, y = cos_7)[name = string("op_1319")];
            tensor<fp16, [1, 32, 64, 32]> var_1320 = mul(x = x1_21, y = sin_7)[name = string("op_1320")];
            tensor<fp16, [1, 32, 64, 32]> var_1321 = add(x = var_1319, y = var_1320)[name = string("op_1321")];
            bool rotated_21_interleave_0 = const()[name = string("rotated_21_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 32, 64, 64]> rotated_21 = concat(axis = var_44, interleave = rotated_21_interleave_0, values = (var_1318, var_1321))[name = string("rotated_21")];
            tensor<int32, [4]> x1_23_begin_0 = const()[name = string("x1_23_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_23_end_0 = const()[name = string("x1_23_end_0"), val = tensor<int32, [4]>([1, 8, 64, 32])];
            tensor<bool, [4]> x1_23_end_mask_0 = const()[name = string("x1_23_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 64, 64]> x_145 = transpose(perm = var_1294, x = var_1293)[name = string("transpose_18")];
            tensor<fp16, [1, 8, 64, 32]> x1_23 = slice_by_index(begin = x1_23_begin_0, end = x1_23_end_0, end_mask = x1_23_end_mask_0, x = x_145)[name = string("x1_23")];
            tensor<int32, [4]> x2_23_begin_0 = const()[name = string("x2_23_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_23_end_0 = const()[name = string("x2_23_end_0"), val = tensor<int32, [4]>([1, 8, 64, 64])];
            tensor<bool, [4]> x2_23_end_mask_0 = const()[name = string("x2_23_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 64, 32]> x2_23 = slice_by_index(begin = x2_23_begin_0, end = x2_23_end_0, end_mask = x2_23_end_mask_0, x = x_145)[name = string("x2_23")];
            tensor<fp16, [1, 8, 64, 32]> var_1337 = mul(x = x1_23, y = cos_7)[name = string("op_1337")];
            tensor<fp16, [1, 8, 64, 32]> var_1338 = mul(x = x2_23, y = sin_7)[name = string("op_1338")];
            tensor<fp16, [1, 8, 64, 32]> var_1339 = sub(x = var_1337, y = var_1338)[name = string("op_1339")];
            tensor<fp16, [1, 8, 64, 32]> var_1340 = mul(x = x2_23, y = cos_7)[name = string("op_1340")];
            tensor<fp16, [1, 8, 64, 32]> var_1341 = mul(x = x1_23, y = sin_7)[name = string("op_1341")];
            tensor<fp16, [1, 8, 64, 32]> var_1342 = add(x = var_1340, y = var_1341)[name = string("op_1342")];
            bool rotated_23_interleave_0 = const()[name = string("rotated_23_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 8, 64, 64]> rotated_23 = concat(axis = var_44, interleave = rotated_23_interleave_0, values = (var_1339, var_1342))[name = string("rotated_23")];
            tensor<int32, [1]> expand_dims_60 = const()[name = string("expand_dims_60"), val = tensor<int32, [1]>([5])];
            tensor<int32, [1]> expand_dims_61 = const()[name = string("expand_dims_61"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_63 = const()[name = string("expand_dims_63"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_64 = const()[name = string("expand_dims_64"), val = tensor<int32, [1]>([6])];
            int32 concat_92_axis_0 = const()[name = string("concat_92_axis_0"), val = int32(0)];
            bool concat_92_interleave_0 = const()[name = string("concat_92_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_92 = concat(axis = concat_92_axis_0, interleave = concat_92_interleave_0, values = (expand_dims_60, expand_dims_61, current_pos, expand_dims_63))[name = string("concat_92")];
            tensor<int32, [1]> concat_93_values1_0 = const()[name = string("concat_93_values1_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> concat_93_values3_0 = const()[name = string("concat_93_values3_0"), val = tensor<int32, [1]>([0])];
            int32 concat_93_axis_0 = const()[name = string("concat_93_axis_0"), val = int32(0)];
            bool concat_93_interleave_0 = const()[name = string("concat_93_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_93 = concat(axis = concat_93_axis_0, interleave = concat_93_interleave_0, values = (expand_dims_64, concat_93_values1_0, var_366, concat_93_values3_0))[name = string("concat_93")];
            tensor<int32, [4]> model_model_kv_cache_0_internal_tensor_assign_11_stride_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_11_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_11_begin_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_11_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_11_end_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_11_end_mask_0"), val = tensor<bool, [4]>([false, true, false, true])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_11_squeeze_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_11_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<fp16, [32, 8, 2048, 64]> model_model_kv_cache_0_internal_tensor_assign_11_cast_fp16 = slice_update(begin = concat_92, begin_mask = model_model_kv_cache_0_internal_tensor_assign_11_begin_mask_0, end = concat_93, end_mask = model_model_kv_cache_0_internal_tensor_assign_11_end_mask_0, squeeze_mask = model_model_kv_cache_0_internal_tensor_assign_11_squeeze_mask_0, stride = model_model_kv_cache_0_internal_tensor_assign_11_stride_0, update = rotated_23, x = coreml_update_state_25)[name = string("model_model_kv_cache_0_internal_tensor_assign_11_cast_fp16")];
            write_state(data = model_model_kv_cache_0_internal_tensor_assign_11_cast_fp16, input = model_model_kv_cache_0)[name = string("coreml_update_state_26_write_state")];
            tensor<fp16, [32, 8, 2048, 64]> coreml_update_state_26 = read_state(input = model_model_kv_cache_0)[name = string("coreml_update_state_26")];
            tensor<int32, [1]> expand_dims_66 = const()[name = string("expand_dims_66"), val = tensor<int32, [1]>([21])];
            tensor<int32, [1]> expand_dims_67 = const()[name = string("expand_dims_67"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_69 = const()[name = string("expand_dims_69"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_70 = const()[name = string("expand_dims_70"), val = tensor<int32, [1]>([22])];
            int32 concat_96_axis_0 = const()[name = string("concat_96_axis_0"), val = int32(0)];
            bool concat_96_interleave_0 = const()[name = string("concat_96_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_96 = concat(axis = concat_96_axis_0, interleave = concat_96_interleave_0, values = (expand_dims_66, expand_dims_67, current_pos, expand_dims_69))[name = string("concat_96")];
            tensor<int32, [1]> concat_97_values1_0 = const()[name = string("concat_97_values1_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> concat_97_values3_0 = const()[name = string("concat_97_values3_0"), val = tensor<int32, [1]>([0])];
            int32 concat_97_axis_0 = const()[name = string("concat_97_axis_0"), val = int32(0)];
            bool concat_97_interleave_0 = const()[name = string("concat_97_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_97 = concat(axis = concat_97_axis_0, interleave = concat_97_interleave_0, values = (expand_dims_70, concat_97_values1_0, var_366, concat_97_values3_0))[name = string("concat_97")];
            tensor<int32, [4]> model_model_kv_cache_0_internal_tensor_assign_12_stride_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_12_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_12_begin_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_12_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_12_end_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_12_end_mask_0"), val = tensor<bool, [4]>([false, true, false, true])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_12_squeeze_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_12_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<fp16, [1, 8, 64, 64]> value_states_33 = transpose(perm = var_1298, x = var_1297)[name = string("transpose_17")];
            tensor<fp16, [32, 8, 2048, 64]> model_model_kv_cache_0_internal_tensor_assign_12_cast_fp16 = slice_update(begin = concat_96, begin_mask = model_model_kv_cache_0_internal_tensor_assign_12_begin_mask_0, end = concat_97, end_mask = model_model_kv_cache_0_internal_tensor_assign_12_end_mask_0, squeeze_mask = model_model_kv_cache_0_internal_tensor_assign_12_squeeze_mask_0, stride = model_model_kv_cache_0_internal_tensor_assign_12_stride_0, update = value_states_33, x = coreml_update_state_26)[name = string("model_model_kv_cache_0_internal_tensor_assign_12_cast_fp16")];
            write_state(data = model_model_kv_cache_0_internal_tensor_assign_12_cast_fp16, input = model_model_kv_cache_0)[name = string("coreml_update_state_27_write_state")];
            tensor<fp16, [32, 8, 2048, 64]> coreml_update_state_27 = read_state(input = model_model_kv_cache_0)[name = string("coreml_update_state_27")];
            tensor<int32, [4]> var_1365_begin_0 = const()[name = string("op_1365_begin_0"), val = tensor<int32, [4]>([5, 0, 0, 0])];
            tensor<int32, [4]> var_1365_end_0 = const()[name = string("op_1365_end_0"), val = tensor<int32, [4]>([6, 8, 2048, 64])];
            tensor<bool, [4]> var_1365_end_mask_0 = const()[name = string("op_1365_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<fp16, [1, 8, 2048, 64]> var_1365_cast_fp16 = slice_by_index(begin = var_1365_begin_0, end = var_1365_end_0, end_mask = var_1365_end_mask_0, x = coreml_update_state_27)[name = string("op_1365_cast_fp16")];
            tensor<int32, [1]> K_layer_cache_11_axes_0 = const()[name = string("K_layer_cache_11_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<fp16, [8, 2048, 64]> K_layer_cache_11_cast_fp16 = squeeze(axes = K_layer_cache_11_axes_0, x = var_1365_cast_fp16)[name = string("K_layer_cache_11_cast_fp16")];
            tensor<int32, [4]> var_1367_begin_0 = const()[name = string("op_1367_begin_0"), val = tensor<int32, [4]>([21, 0, 0, 0])];
            tensor<int32, [4]> var_1367_end_0 = const()[name = string("op_1367_end_0"), val = tensor<int32, [4]>([22, 8, 2048, 64])];
            tensor<bool, [4]> var_1367_end_mask_0 = const()[name = string("op_1367_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<fp16, [1, 8, 2048, 64]> var_1367_cast_fp16 = slice_by_index(begin = var_1367_begin_0, end = var_1367_end_0, end_mask = var_1367_end_mask_0, x = coreml_update_state_27)[name = string("op_1367_cast_fp16")];
            tensor<int32, [1]> V_layer_cache_11_axes_0 = const()[name = string("V_layer_cache_11_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<fp16, [8, 2048, 64]> V_layer_cache_11_cast_fp16 = squeeze(axes = V_layer_cache_11_axes_0, x = var_1367_cast_fp16)[name = string("V_layer_cache_11_cast_fp16")];
            tensor<int32, [1]> x_151_axes_0 = const()[name = string("x_151_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [8, 1, 2048, 64]> x_151_cast_fp16 = expand_dims(axes = x_151_axes_0, x = K_layer_cache_11_cast_fp16)[name = string("x_151_cast_fp16")];
            tensor<int32, [4]> var_1376 = const()[name = string("op_1376"), val = tensor<int32, [4]>([1, 4, 1, 1])];
            tensor<fp16, [8, 4, 2048, 64]> x_153_cast_fp16 = tile(reps = var_1376, x = x_151_cast_fp16)[name = string("x_153_cast_fp16")];
            tensor<int32, [4]> var_1380 = const()[name = string("op_1380"), val = tensor<int32, [4]>([1, -1, 2048, 64])];
            tensor<fp16, [1, 32, 2048, 64]> var_1381_cast_fp16 = reshape(shape = var_1380, x = x_153_cast_fp16)[name = string("op_1381_cast_fp16")];
            tensor<int32, [1]> x_157_axes_0 = const()[name = string("x_157_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [8, 1, 2048, 64]> x_157_cast_fp16 = expand_dims(axes = x_157_axes_0, x = V_layer_cache_11_cast_fp16)[name = string("x_157_cast_fp16")];
            tensor<int32, [4]> var_1383 = const()[name = string("op_1383"), val = tensor<int32, [4]>([1, 4, 1, 1])];
            tensor<fp16, [8, 4, 2048, 64]> x_159_cast_fp16 = tile(reps = var_1383, x = x_157_cast_fp16)[name = string("x_159_cast_fp16")];
            bool var_1390_transpose_x_0 = const()[name = string("op_1390_transpose_x_0"), val = bool(false)];
            bool var_1390_transpose_y_0 = const()[name = string("op_1390_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 32, 64, 2048]> var_1390_cast_fp16 = matmul(transpose_x = var_1390_transpose_x_0, transpose_y = var_1390_transpose_y_0, x = rotated_21, y = var_1381_cast_fp16)[name = string("op_1390_cast_fp16")];
            fp16 var_1391_to_fp16 = const()[name = string("op_1391_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 32, 64, 2048]> attn_weights_11_cast_fp16 = mul(x = var_1390_cast_fp16, y = var_1391_to_fp16)[name = string("attn_weights_11_cast_fp16")];
            tensor<fp16, [1, 32, 64, 2048]> x_161_cast_fp16 = add(x = attn_weights_11_cast_fp16, y = causal_mask)[name = string("x_161_cast_fp16")];
            tensor<int32, [1]> reduce_max_5_axes_0 = const()[name = string("reduce_max_5_axes_0"), val = tensor<int32, [1]>([-1])];
            bool reduce_max_5_keep_dims_0 = const()[name = string("reduce_max_5_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 64, 1]> reduce_max_5_cast_fp16 = reduce_max(axes = reduce_max_5_axes_0, keep_dims = reduce_max_5_keep_dims_0, x = x_161_cast_fp16)[name = string("reduce_max_5_cast_fp16")];
            tensor<fp16, [1, 32, 64, 2048]> x_163_cast_fp16 = sub(x = x_161_cast_fp16, y = reduce_max_5_cast_fp16)[name = string("x_163_cast_fp16")];
            tensor<fp16, [1, 32, 64, 2048]> exp_x_11_cast_fp16 = exp(x = x_163_cast_fp16)[name = string("exp_x_11_cast_fp16")];
            tensor<int32, [1]> var_1402_axes_0 = const()[name = string("op_1402_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1402_keep_dims_0 = const()[name = string("op_1402_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 64, 1]> var_1402_cast_fp16 = reduce_sum(axes = var_1402_axes_0, keep_dims = var_1402_keep_dims_0, x = exp_x_11_cast_fp16)[name = string("op_1402_cast_fp16")];
            tensor<fp16, [1, 32, 64, 2048]> var_1403_cast_fp16 = real_div(x = exp_x_11_cast_fp16, y = var_1402_cast_fp16)[name = string("op_1403_cast_fp16")];
            tensor<int32, [3]> concat_102 = const()[name = string("concat_102"), val = tensor<int32, [3]>([32, 64, 2048])];
            tensor<fp16, [32, 64, 2048]> reshape_15_cast_fp16 = reshape(shape = concat_102, x = var_1403_cast_fp16)[name = string("reshape_15_cast_fp16")];
            tensor<int32, [3]> concat_103 = const()[name = string("concat_103"), val = tensor<int32, [3]>([32, 2048, 64])];
            tensor<fp16, [32, 2048, 64]> reshape_16_cast_fp16 = reshape(shape = concat_103, x = x_159_cast_fp16)[name = string("reshape_16_cast_fp16")];
            bool matmul_5_transpose_x_0 = const()[name = string("matmul_5_transpose_x_0"), val = bool(false)];
            bool matmul_5_transpose_y_0 = const()[name = string("matmul_5_transpose_y_0"), val = bool(false)];
            tensor<fp16, [32, 64, 64]> matmul_5_cast_fp16 = matmul(transpose_x = matmul_5_transpose_x_0, transpose_y = matmul_5_transpose_y_0, x = reshape_15_cast_fp16, y = reshape_16_cast_fp16)[name = string("matmul_5_cast_fp16")];
            tensor<int32, [4]> concat_107 = const()[name = string("concat_107"), val = tensor<int32, [4]>([1, 32, 64, 64])];
            tensor<fp16, [1, 32, 64, 64]> reshape_17_cast_fp16 = reshape(shape = concat_107, x = matmul_5_cast_fp16)[name = string("reshape_17_cast_fp16")];
            tensor<int32, [4]> var_1406_perm_0 = const()[name = string("op_1406_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1408 = const()[name = string("op_1408"), val = tensor<int32, [3]>([1, 64, 2048])];
            tensor<fp16, [1, 64, 32, 64]> var_1406_cast_fp16 = transpose(perm = var_1406_perm_0, x = reshape_17_cast_fp16)[name = string("transpose_16")];
            tensor<fp16, [1, 64, 2048]> input_75_cast_fp16 = reshape(shape = var_1408, x = var_1406_cast_fp16)[name = string("input_75_cast_fp16")];
            tensor<fp16, [2048, 2048]> model_model_layers_5_self_attn_o_proj_weight_promoted_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271318592))), lut = tensor<fp16, [256, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(273415808))))[name = string("model_model_layers_5_self_attn_o_proj_weight_promoted_to_fp16_palettized")];
            tensor<fp16, [1, 64, 2048]> linear_5_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_5_self_attn_o_proj_weight_promoted_to_fp16_palettized, x = input_75_cast_fp16)[name = string("linear_5_cast_fp16")];
            tensor<fp16, [1, 64, 2048]> hidden_states_45_cast_fp16 = add(x = hidden_states_41_cast_fp16, y = linear_5_cast_fp16)[name = string("hidden_states_45_cast_fp16")];
            tensor<int32, [1]> mean_23_axes_0 = const()[name = string("mean_23_axes_0"), val = tensor<int32, [1]>([-1])];
            bool mean_23_keep_dims_0 = const()[name = string("mean_23_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 64, 1]> mean_23_cast_fp16 = reduce_mean(axes = mean_23_axes_0, keep_dims = mean_23_keep_dims_0, x = hidden_states_45_cast_fp16)[name = string("mean_23_cast_fp16")];
            tensor<fp16, [1, 64, 2048]> input_77_cast_fp16 = sub(x = hidden_states_45_cast_fp16, y = mean_23_cast_fp16)[name = string("input_77_cast_fp16")];
            tensor<int32, [1]> var_1419_axes_0 = const()[name = string("op_1419_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2048]> model_model_layers_5_post_attention_layernorm_weight_to_fp16 = const()[name = string("model_model_layers_5_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(273424064)))];
            tensor<fp16, [1, 64, 2048]> var_1419_cast_fp16 = layer_norm(axes = var_1419_axes_0, epsilon = var_46_to_fp16, gamma = model_model_layers_5_post_attention_layernorm_weight_to_fp16, x = input_77_cast_fp16)[name = string("op_1419_cast_fp16")];
            tensor<int32, [3]> var_1426 = const()[name = string("op_1426"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> input_79_axes_0 = const()[name = string("input_79_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 64]> var_1427 = transpose(perm = var_1426, x = var_1419_cast_fp16)[name = string("transpose_15")];
            tensor<fp16, [1, 2048, 1, 64]> input_79 = expand_dims(axes = input_79_axes_0, x = var_1427)[name = string("input_79")];
            string input_81_pad_type_0 = const()[name = string("input_81_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_81_strides_0 = const()[name = string("input_81_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_81_pad_0 = const()[name = string("input_81_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_81_dilations_0 = const()[name = string("input_81_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_81_groups_0 = const()[name = string("input_81_groups_0"), val = int32(1)];
            tensor<fp16, [1, 8192, 1, 64]> input_81 = conv(dilations = input_81_dilations_0, groups = input_81_groups_0, pad = input_81_pad_0, pad_type = input_81_pad_type_0, strides = input_81_strides_0, weight = model_model_layers_5_mlp_gate_proj_weight_palettized, x = input_79)[name = string("input_81")];
            string up_states_11_pad_type_0 = const()[name = string("up_states_11_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> up_states_11_strides_0 = const()[name = string("up_states_11_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> up_states_11_pad_0 = const()[name = string("up_states_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> up_states_11_dilations_0 = const()[name = string("up_states_11_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 up_states_11_groups_0 = const()[name = string("up_states_11_groups_0"), val = int32(1)];
            tensor<fp16, [1, 8192, 1, 64]> up_states_11 = conv(dilations = up_states_11_dilations_0, groups = up_states_11_groups_0, pad = up_states_11_pad_0, pad_type = up_states_11_pad_type_0, strides = up_states_11_strides_0, weight = model_model_layers_5_mlp_up_proj_weight_palettized, x = input_79)[name = string("up_states_11")];
            tensor<fp16, [1, 8192, 1, 64]> gate_states_11 = silu(x = input_81)[name = string("gate_states_11")];
            tensor<fp16, [1, 8192, 1, 64]> input_83 = mul(x = gate_states_11, y = up_states_11)[name = string("input_83")];
            string hidden_states_47_pad_type_0 = const()[name = string("hidden_states_47_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_47_strides_0 = const()[name = string("hidden_states_47_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_47_pad_0 = const()[name = string("hidden_states_47_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_47_dilations_0 = const()[name = string("hidden_states_47_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_47_groups_0 = const()[name = string("hidden_states_47_groups_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 1, 64]> hidden_states_47 = conv(dilations = hidden_states_47_dilations_0, groups = hidden_states_47_groups_0, pad = hidden_states_47_pad_0, pad_type = hidden_states_47_pad_type_0, strides = hidden_states_47_strides_0, weight = model_model_layers_5_mlp_down_proj_weight_palettized, x = input_83)[name = string("hidden_states_47")];
            tensor<int32, [1]> var_1449_axes_0 = const()[name = string("op_1449_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 64]> var_1449 = squeeze(axes = var_1449_axes_0, x = hidden_states_47)[name = string("op_1449")];
            tensor<int32, [3]> var_1450 = const()[name = string("op_1450"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 64, 2048]> var_1451 = transpose(perm = var_1450, x = var_1449)[name = string("transpose_14")];
            tensor<fp16, [1, 64, 2048]> hidden_states_49_cast_fp16 = add(x = hidden_states_45_cast_fp16, y = var_1451)[name = string("hidden_states_49_cast_fp16")];
            tensor<int32, [1]> mean_25_axes_0 = const()[name = string("mean_25_axes_0"), val = tensor<int32, [1]>([-1])];
            bool mean_25_keep_dims_0 = const()[name = string("mean_25_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 64, 1]> mean_25_cast_fp16 = reduce_mean(axes = mean_25_axes_0, keep_dims = mean_25_keep_dims_0, x = hidden_states_49_cast_fp16)[name = string("mean_25_cast_fp16")];
            tensor<fp16, [1, 64, 2048]> input_85_cast_fp16 = sub(x = hidden_states_49_cast_fp16, y = mean_25_cast_fp16)[name = string("input_85_cast_fp16")];
            tensor<int32, [1]> var_1459_axes_0 = const()[name = string("op_1459_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2048]> model_model_layers_6_input_layernorm_weight_to_fp16 = const()[name = string("model_model_layers_6_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(273428224)))];
            tensor<fp16, [1, 64, 2048]> var_1459_cast_fp16 = layer_norm(axes = var_1459_axes_0, epsilon = var_46_to_fp16, gamma = model_model_layers_6_input_layernorm_weight_to_fp16, x = input_85_cast_fp16)[name = string("op_1459_cast_fp16")];
            tensor<int32, [3]> var_1463 = const()[name = string("op_1463"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1465_axes_0 = const()[name = string("op_1465_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 64]> var_1464 = transpose(perm = var_1463, x = var_1459_cast_fp16)[name = string("transpose_13")];
            tensor<fp16, [1, 2048, 1, 64]> var_1465 = expand_dims(axes = var_1465_axes_0, x = var_1464)[name = string("op_1465")];
            string query_states_25_pad_type_0 = const()[name = string("query_states_25_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_states_25_strides_0 = const()[name = string("query_states_25_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_states_25_pad_0 = const()[name = string("query_states_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_states_25_dilations_0 = const()[name = string("query_states_25_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_states_25_groups_0 = const()[name = string("query_states_25_groups_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 1, 64]> query_states_25 = conv(dilations = query_states_25_dilations_0, groups = query_states_25_groups_0, pad = query_states_25_pad_0, pad_type = query_states_25_pad_type_0, strides = query_states_25_strides_0, weight = model_model_layers_6_self_attn_q_proj_weight_palettized, x = var_1465)[name = string("query_states_25")];
            string key_states_37_pad_type_0 = const()[name = string("key_states_37_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_states_37_strides_0 = const()[name = string("key_states_37_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_states_37_pad_0 = const()[name = string("key_states_37_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_states_37_dilations_0 = const()[name = string("key_states_37_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_states_37_groups_0 = const()[name = string("key_states_37_groups_0"), val = int32(1)];
            tensor<fp16, [1, 512, 1, 64]> key_states_37 = conv(dilations = key_states_37_dilations_0, groups = key_states_37_groups_0, pad = key_states_37_pad_0, pad_type = key_states_37_pad_type_0, strides = key_states_37_strides_0, weight = model_model_layers_6_self_attn_k_proj_weight_palettized, x = var_1465)[name = string("key_states_37")];
            string value_states_37_pad_type_0 = const()[name = string("value_states_37_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_states_37_strides_0 = const()[name = string("value_states_37_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_states_37_pad_0 = const()[name = string("value_states_37_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_states_37_dilations_0 = const()[name = string("value_states_37_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_states_37_groups_0 = const()[name = string("value_states_37_groups_0"), val = int32(1)];
            tensor<fp16, [1, 512, 1, 64]> value_states_37 = conv(dilations = value_states_37_dilations_0, groups = value_states_37_groups_0, pad = value_states_37_pad_0, pad_type = value_states_37_pad_type_0, strides = value_states_37_strides_0, weight = model_model_layers_6_self_attn_v_proj_weight_palettized, x = var_1465)[name = string("value_states_37")];
            tensor<int32, [4]> var_1485 = const()[name = string("op_1485"), val = tensor<int32, [4]>([1, 32, 64, 64])];
            tensor<fp16, [1, 32, 64, 64]> var_1486 = reshape(shape = var_1485, x = query_states_25)[name = string("op_1486")];
            tensor<int32, [4]> var_1487 = const()[name = string("op_1487"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_1489 = const()[name = string("op_1489"), val = tensor<int32, [4]>([1, 8, 64, 64])];
            tensor<fp16, [1, 8, 64, 64]> var_1490 = reshape(shape = var_1489, x = key_states_37)[name = string("op_1490")];
            tensor<int32, [4]> var_1491 = const()[name = string("op_1491"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_1493 = const()[name = string("op_1493"), val = tensor<int32, [4]>([1, 8, 64, 64])];
            tensor<fp16, [1, 8, 64, 64]> var_1494 = reshape(shape = var_1493, x = value_states_37)[name = string("op_1494")];
            tensor<int32, [4]> var_1495 = const()[name = string("op_1495"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> x1_25_begin_0 = const()[name = string("x1_25_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_25_end_0 = const()[name = string("x1_25_end_0"), val = tensor<int32, [4]>([1, 32, 64, 32])];
            tensor<bool, [4]> x1_25_end_mask_0 = const()[name = string("x1_25_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 64, 64]> x_169 = transpose(perm = var_1487, x = var_1486)[name = string("transpose_12")];
            tensor<fp16, [1, 32, 64, 32]> x1_25 = slice_by_index(begin = x1_25_begin_0, end = x1_25_end_0, end_mask = x1_25_end_mask_0, x = x_169)[name = string("x1_25")];
            tensor<int32, [4]> x2_25_begin_0 = const()[name = string("x2_25_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_25_end_0 = const()[name = string("x2_25_end_0"), val = tensor<int32, [4]>([1, 32, 64, 64])];
            tensor<bool, [4]> x2_25_end_mask_0 = const()[name = string("x2_25_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 64, 32]> x2_25 = slice_by_index(begin = x2_25_begin_0, end = x2_25_end_0, end_mask = x2_25_end_mask_0, x = x_169)[name = string("x2_25")];
            tensor<fp16, [1, 32, 64, 32]> var_1513 = mul(x = x1_25, y = cos_7)[name = string("op_1513")];
            tensor<fp16, [1, 32, 64, 32]> var_1514 = mul(x = x2_25, y = sin_7)[name = string("op_1514")];
            tensor<fp16, [1, 32, 64, 32]> var_1515 = sub(x = var_1513, y = var_1514)[name = string("op_1515")];
            tensor<fp16, [1, 32, 64, 32]> var_1516 = mul(x = x2_25, y = cos_7)[name = string("op_1516")];
            tensor<fp16, [1, 32, 64, 32]> var_1517 = mul(x = x1_25, y = sin_7)[name = string("op_1517")];
            tensor<fp16, [1, 32, 64, 32]> var_1518 = add(x = var_1516, y = var_1517)[name = string("op_1518")];
            bool rotated_25_interleave_0 = const()[name = string("rotated_25_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 32, 64, 64]> rotated_25 = concat(axis = var_44, interleave = rotated_25_interleave_0, values = (var_1515, var_1518))[name = string("rotated_25")];
            tensor<int32, [4]> x1_27_begin_0 = const()[name = string("x1_27_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_27_end_0 = const()[name = string("x1_27_end_0"), val = tensor<int32, [4]>([1, 8, 64, 32])];
            tensor<bool, [4]> x1_27_end_mask_0 = const()[name = string("x1_27_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 64, 64]> x_173 = transpose(perm = var_1491, x = var_1490)[name = string("transpose_11")];
            tensor<fp16, [1, 8, 64, 32]> x1_27 = slice_by_index(begin = x1_27_begin_0, end = x1_27_end_0, end_mask = x1_27_end_mask_0, x = x_173)[name = string("x1_27")];
            tensor<int32, [4]> x2_27_begin_0 = const()[name = string("x2_27_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_27_end_0 = const()[name = string("x2_27_end_0"), val = tensor<int32, [4]>([1, 8, 64, 64])];
            tensor<bool, [4]> x2_27_end_mask_0 = const()[name = string("x2_27_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 64, 32]> x2_27 = slice_by_index(begin = x2_27_begin_0, end = x2_27_end_0, end_mask = x2_27_end_mask_0, x = x_173)[name = string("x2_27")];
            tensor<fp16, [1, 8, 64, 32]> var_1534 = mul(x = x1_27, y = cos_7)[name = string("op_1534")];
            tensor<fp16, [1, 8, 64, 32]> var_1535 = mul(x = x2_27, y = sin_7)[name = string("op_1535")];
            tensor<fp16, [1, 8, 64, 32]> var_1536 = sub(x = var_1534, y = var_1535)[name = string("op_1536")];
            tensor<fp16, [1, 8, 64, 32]> var_1537 = mul(x = x2_27, y = cos_7)[name = string("op_1537")];
            tensor<fp16, [1, 8, 64, 32]> var_1538 = mul(x = x1_27, y = sin_7)[name = string("op_1538")];
            tensor<fp16, [1, 8, 64, 32]> var_1539 = add(x = var_1537, y = var_1538)[name = string("op_1539")];
            bool rotated_27_interleave_0 = const()[name = string("rotated_27_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 8, 64, 64]> rotated_27 = concat(axis = var_44, interleave = rotated_27_interleave_0, values = (var_1536, var_1539))[name = string("rotated_27")];
            tensor<int32, [1]> expand_dims_72 = const()[name = string("expand_dims_72"), val = tensor<int32, [1]>([6])];
            tensor<int32, [1]> expand_dims_73 = const()[name = string("expand_dims_73"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_75 = const()[name = string("expand_dims_75"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_76 = const()[name = string("expand_dims_76"), val = tensor<int32, [1]>([7])];
            int32 concat_110_axis_0 = const()[name = string("concat_110_axis_0"), val = int32(0)];
            bool concat_110_interleave_0 = const()[name = string("concat_110_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_110 = concat(axis = concat_110_axis_0, interleave = concat_110_interleave_0, values = (expand_dims_72, expand_dims_73, current_pos, expand_dims_75))[name = string("concat_110")];
            tensor<int32, [1]> concat_111_values1_0 = const()[name = string("concat_111_values1_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> concat_111_values3_0 = const()[name = string("concat_111_values3_0"), val = tensor<int32, [1]>([0])];
            int32 concat_111_axis_0 = const()[name = string("concat_111_axis_0"), val = int32(0)];
            bool concat_111_interleave_0 = const()[name = string("concat_111_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_111 = concat(axis = concat_111_axis_0, interleave = concat_111_interleave_0, values = (expand_dims_76, concat_111_values1_0, var_366, concat_111_values3_0))[name = string("concat_111")];
            tensor<int32, [4]> model_model_kv_cache_0_internal_tensor_assign_13_stride_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_13_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_13_begin_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_13_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_13_end_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_13_end_mask_0"), val = tensor<bool, [4]>([false, true, false, true])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_13_squeeze_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_13_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<fp16, [32, 8, 2048, 64]> model_model_kv_cache_0_internal_tensor_assign_13_cast_fp16 = slice_update(begin = concat_110, begin_mask = model_model_kv_cache_0_internal_tensor_assign_13_begin_mask_0, end = concat_111, end_mask = model_model_kv_cache_0_internal_tensor_assign_13_end_mask_0, squeeze_mask = model_model_kv_cache_0_internal_tensor_assign_13_squeeze_mask_0, stride = model_model_kv_cache_0_internal_tensor_assign_13_stride_0, update = rotated_27, x = coreml_update_state_27)[name = string("model_model_kv_cache_0_internal_tensor_assign_13_cast_fp16")];
            write_state(data = model_model_kv_cache_0_internal_tensor_assign_13_cast_fp16, input = model_model_kv_cache_0)[name = string("coreml_update_state_28_write_state")];
            tensor<fp16, [32, 8, 2048, 64]> coreml_update_state_28 = read_state(input = model_model_kv_cache_0)[name = string("coreml_update_state_28")];
            tensor<int32, [1]> expand_dims_78 = const()[name = string("expand_dims_78"), val = tensor<int32, [1]>([22])];
            tensor<int32, [1]> expand_dims_79 = const()[name = string("expand_dims_79"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_81 = const()[name = string("expand_dims_81"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_82 = const()[name = string("expand_dims_82"), val = tensor<int32, [1]>([23])];
            int32 concat_114_axis_0 = const()[name = string("concat_114_axis_0"), val = int32(0)];
            bool concat_114_interleave_0 = const()[name = string("concat_114_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_114 = concat(axis = concat_114_axis_0, interleave = concat_114_interleave_0, values = (expand_dims_78, expand_dims_79, current_pos, expand_dims_81))[name = string("concat_114")];
            tensor<int32, [1]> concat_115_values1_0 = const()[name = string("concat_115_values1_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> concat_115_values3_0 = const()[name = string("concat_115_values3_0"), val = tensor<int32, [1]>([0])];
            int32 concat_115_axis_0 = const()[name = string("concat_115_axis_0"), val = int32(0)];
            bool concat_115_interleave_0 = const()[name = string("concat_115_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_115 = concat(axis = concat_115_axis_0, interleave = concat_115_interleave_0, values = (expand_dims_82, concat_115_values1_0, var_366, concat_115_values3_0))[name = string("concat_115")];
            tensor<int32, [4]> model_model_kv_cache_0_internal_tensor_assign_14_stride_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_14_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_14_begin_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_14_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_14_end_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_14_end_mask_0"), val = tensor<bool, [4]>([false, true, false, true])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_14_squeeze_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_14_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<fp16, [1, 8, 64, 64]> value_states_39 = transpose(perm = var_1495, x = var_1494)[name = string("transpose_10")];
            tensor<fp16, [32, 8, 2048, 64]> model_model_kv_cache_0_internal_tensor_assign_14_cast_fp16 = slice_update(begin = concat_114, begin_mask = model_model_kv_cache_0_internal_tensor_assign_14_begin_mask_0, end = concat_115, end_mask = model_model_kv_cache_0_internal_tensor_assign_14_end_mask_0, squeeze_mask = model_model_kv_cache_0_internal_tensor_assign_14_squeeze_mask_0, stride = model_model_kv_cache_0_internal_tensor_assign_14_stride_0, update = value_states_39, x = coreml_update_state_28)[name = string("model_model_kv_cache_0_internal_tensor_assign_14_cast_fp16")];
            write_state(data = model_model_kv_cache_0_internal_tensor_assign_14_cast_fp16, input = model_model_kv_cache_0)[name = string("coreml_update_state_29_write_state")];
            tensor<fp16, [32, 8, 2048, 64]> coreml_update_state_29 = read_state(input = model_model_kv_cache_0)[name = string("coreml_update_state_29")];
            tensor<int32, [4]> var_1562_begin_0 = const()[name = string("op_1562_begin_0"), val = tensor<int32, [4]>([6, 0, 0, 0])];
            tensor<int32, [4]> var_1562_end_0 = const()[name = string("op_1562_end_0"), val = tensor<int32, [4]>([7, 8, 2048, 64])];
            tensor<bool, [4]> var_1562_end_mask_0 = const()[name = string("op_1562_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<fp16, [1, 8, 2048, 64]> var_1562_cast_fp16 = slice_by_index(begin = var_1562_begin_0, end = var_1562_end_0, end_mask = var_1562_end_mask_0, x = coreml_update_state_29)[name = string("op_1562_cast_fp16")];
            tensor<int32, [1]> K_layer_cache_13_axes_0 = const()[name = string("K_layer_cache_13_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<fp16, [8, 2048, 64]> K_layer_cache_13_cast_fp16 = squeeze(axes = K_layer_cache_13_axes_0, x = var_1562_cast_fp16)[name = string("K_layer_cache_13_cast_fp16")];
            tensor<int32, [4]> var_1564_begin_0 = const()[name = string("op_1564_begin_0"), val = tensor<int32, [4]>([22, 0, 0, 0])];
            tensor<int32, [4]> var_1564_end_0 = const()[name = string("op_1564_end_0"), val = tensor<int32, [4]>([23, 8, 2048, 64])];
            tensor<bool, [4]> var_1564_end_mask_0 = const()[name = string("op_1564_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<fp16, [1, 8, 2048, 64]> var_1564_cast_fp16 = slice_by_index(begin = var_1564_begin_0, end = var_1564_end_0, end_mask = var_1564_end_mask_0, x = coreml_update_state_29)[name = string("op_1564_cast_fp16")];
            tensor<int32, [1]> V_layer_cache_13_axes_0 = const()[name = string("V_layer_cache_13_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<fp16, [8, 2048, 64]> V_layer_cache_13_cast_fp16 = squeeze(axes = V_layer_cache_13_axes_0, x = var_1564_cast_fp16)[name = string("V_layer_cache_13_cast_fp16")];
            tensor<int32, [1]> x_179_axes_0 = const()[name = string("x_179_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [8, 1, 2048, 64]> x_179_cast_fp16 = expand_dims(axes = x_179_axes_0, x = K_layer_cache_13_cast_fp16)[name = string("x_179_cast_fp16")];
            tensor<int32, [4]> var_1573 = const()[name = string("op_1573"), val = tensor<int32, [4]>([1, 4, 1, 1])];
            tensor<fp16, [8, 4, 2048, 64]> x_181_cast_fp16 = tile(reps = var_1573, x = x_179_cast_fp16)[name = string("x_181_cast_fp16")];
            tensor<int32, [4]> var_1577 = const()[name = string("op_1577"), val = tensor<int32, [4]>([1, -1, 2048, 64])];
            tensor<fp16, [1, 32, 2048, 64]> var_1578_cast_fp16 = reshape(shape = var_1577, x = x_181_cast_fp16)[name = string("op_1578_cast_fp16")];
            tensor<int32, [1]> x_185_axes_0 = const()[name = string("x_185_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [8, 1, 2048, 64]> x_185_cast_fp16 = expand_dims(axes = x_185_axes_0, x = V_layer_cache_13_cast_fp16)[name = string("x_185_cast_fp16")];
            tensor<int32, [4]> var_1580 = const()[name = string("op_1580"), val = tensor<int32, [4]>([1, 4, 1, 1])];
            tensor<fp16, [8, 4, 2048, 64]> x_187_cast_fp16 = tile(reps = var_1580, x = x_185_cast_fp16)[name = string("x_187_cast_fp16")];
            bool var_1587_transpose_x_0 = const()[name = string("op_1587_transpose_x_0"), val = bool(false)];
            bool var_1587_transpose_y_0 = const()[name = string("op_1587_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 32, 64, 2048]> var_1587_cast_fp16 = matmul(transpose_x = var_1587_transpose_x_0, transpose_y = var_1587_transpose_y_0, x = rotated_25, y = var_1578_cast_fp16)[name = string("op_1587_cast_fp16")];
            fp16 var_1588_to_fp16 = const()[name = string("op_1588_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 32, 64, 2048]> attn_weights_13_cast_fp16 = mul(x = var_1587_cast_fp16, y = var_1588_to_fp16)[name = string("attn_weights_13_cast_fp16")];
            tensor<fp16, [1, 32, 64, 2048]> x_189_cast_fp16 = add(x = attn_weights_13_cast_fp16, y = causal_mask)[name = string("x_189_cast_fp16")];
            tensor<int32, [1]> reduce_max_6_axes_0 = const()[name = string("reduce_max_6_axes_0"), val = tensor<int32, [1]>([-1])];
            bool reduce_max_6_keep_dims_0 = const()[name = string("reduce_max_6_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 64, 1]> reduce_max_6_cast_fp16 = reduce_max(axes = reduce_max_6_axes_0, keep_dims = reduce_max_6_keep_dims_0, x = x_189_cast_fp16)[name = string("reduce_max_6_cast_fp16")];
            tensor<fp16, [1, 32, 64, 2048]> x_191_cast_fp16 = sub(x = x_189_cast_fp16, y = reduce_max_6_cast_fp16)[name = string("x_191_cast_fp16")];
            tensor<fp16, [1, 32, 64, 2048]> exp_x_13_cast_fp16 = exp(x = x_191_cast_fp16)[name = string("exp_x_13_cast_fp16")];
            tensor<int32, [1]> var_1599_axes_0 = const()[name = string("op_1599_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1599_keep_dims_0 = const()[name = string("op_1599_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 64, 1]> var_1599_cast_fp16 = reduce_sum(axes = var_1599_axes_0, keep_dims = var_1599_keep_dims_0, x = exp_x_13_cast_fp16)[name = string("op_1599_cast_fp16")];
            tensor<fp16, [1, 32, 64, 2048]> var_1600_cast_fp16 = real_div(x = exp_x_13_cast_fp16, y = var_1599_cast_fp16)[name = string("op_1600_cast_fp16")];
            tensor<int32, [3]> concat_120 = const()[name = string("concat_120"), val = tensor<int32, [3]>([32, 64, 2048])];
            tensor<fp16, [32, 64, 2048]> reshape_18_cast_fp16 = reshape(shape = concat_120, x = var_1600_cast_fp16)[name = string("reshape_18_cast_fp16")];
            tensor<int32, [3]> concat_121 = const()[name = string("concat_121"), val = tensor<int32, [3]>([32, 2048, 64])];
            tensor<fp16, [32, 2048, 64]> reshape_19_cast_fp16 = reshape(shape = concat_121, x = x_187_cast_fp16)[name = string("reshape_19_cast_fp16")];
            bool matmul_6_transpose_x_0 = const()[name = string("matmul_6_transpose_x_0"), val = bool(false)];
            bool matmul_6_transpose_y_0 = const()[name = string("matmul_6_transpose_y_0"), val = bool(false)];
            tensor<fp16, [32, 64, 64]> matmul_6_cast_fp16 = matmul(transpose_x = matmul_6_transpose_x_0, transpose_y = matmul_6_transpose_y_0, x = reshape_18_cast_fp16, y = reshape_19_cast_fp16)[name = string("matmul_6_cast_fp16")];
            tensor<int32, [4]> concat_125 = const()[name = string("concat_125"), val = tensor<int32, [4]>([1, 32, 64, 64])];
            tensor<fp16, [1, 32, 64, 64]> reshape_20_cast_fp16 = reshape(shape = concat_125, x = matmul_6_cast_fp16)[name = string("reshape_20_cast_fp16")];
            tensor<int32, [4]> var_1603_perm_0 = const()[name = string("op_1603_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1605 = const()[name = string("op_1605"), val = tensor<int32, [3]>([1, 64, 2048])];
            tensor<fp16, [1, 64, 32, 64]> var_1603_cast_fp16 = transpose(perm = var_1603_perm_0, x = reshape_20_cast_fp16)[name = string("transpose_9")];
            tensor<fp16, [1, 64, 2048]> input_89_cast_fp16 = reshape(shape = var_1605, x = var_1603_cast_fp16)[name = string("input_89_cast_fp16")];
            tensor<fp16, [2048, 2048]> model_model_layers_6_self_attn_o_proj_weight_promoted_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(273432384))), lut = tensor<fp16, [256, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275529600))))[name = string("model_model_layers_6_self_attn_o_proj_weight_promoted_to_fp16_palettized")];
            tensor<fp16, [1, 64, 2048]> linear_6_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_6_self_attn_o_proj_weight_promoted_to_fp16_palettized, x = input_89_cast_fp16)[name = string("linear_6_cast_fp16")];
            tensor<fp16, [1, 64, 2048]> hidden_states_53_cast_fp16 = add(x = hidden_states_49_cast_fp16, y = linear_6_cast_fp16)[name = string("hidden_states_53_cast_fp16")];
            tensor<int32, [1]> mean_27_axes_0 = const()[name = string("mean_27_axes_0"), val = tensor<int32, [1]>([-1])];
            bool mean_27_keep_dims_0 = const()[name = string("mean_27_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 64, 1]> mean_27_cast_fp16 = reduce_mean(axes = mean_27_axes_0, keep_dims = mean_27_keep_dims_0, x = hidden_states_53_cast_fp16)[name = string("mean_27_cast_fp16")];
            tensor<fp16, [1, 64, 2048]> input_91_cast_fp16 = sub(x = hidden_states_53_cast_fp16, y = mean_27_cast_fp16)[name = string("input_91_cast_fp16")];
            tensor<int32, [1]> var_1616_axes_0 = const()[name = string("op_1616_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2048]> model_model_layers_6_post_attention_layernorm_weight_to_fp16 = const()[name = string("model_model_layers_6_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275537856)))];
            tensor<fp16, [1, 64, 2048]> var_1616_cast_fp16 = layer_norm(axes = var_1616_axes_0, epsilon = var_46_to_fp16, gamma = model_model_layers_6_post_attention_layernorm_weight_to_fp16, x = input_91_cast_fp16)[name = string("op_1616_cast_fp16")];
            tensor<int32, [3]> var_1623 = const()[name = string("op_1623"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> input_93_axes_0 = const()[name = string("input_93_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 64]> var_1624 = transpose(perm = var_1623, x = var_1616_cast_fp16)[name = string("transpose_8")];
            tensor<fp16, [1, 2048, 1, 64]> input_93 = expand_dims(axes = input_93_axes_0, x = var_1624)[name = string("input_93")];
            string input_95_pad_type_0 = const()[name = string("input_95_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_95_strides_0 = const()[name = string("input_95_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_95_pad_0 = const()[name = string("input_95_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_95_dilations_0 = const()[name = string("input_95_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_95_groups_0 = const()[name = string("input_95_groups_0"), val = int32(1)];
            tensor<fp16, [1, 8192, 1, 64]> input_95 = conv(dilations = input_95_dilations_0, groups = input_95_groups_0, pad = input_95_pad_0, pad_type = input_95_pad_type_0, strides = input_95_strides_0, weight = model_model_layers_6_mlp_gate_proj_weight_palettized, x = input_93)[name = string("input_95")];
            string up_states_13_pad_type_0 = const()[name = string("up_states_13_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> up_states_13_strides_0 = const()[name = string("up_states_13_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> up_states_13_pad_0 = const()[name = string("up_states_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> up_states_13_dilations_0 = const()[name = string("up_states_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 up_states_13_groups_0 = const()[name = string("up_states_13_groups_0"), val = int32(1)];
            tensor<fp16, [1, 8192, 1, 64]> up_states_13 = conv(dilations = up_states_13_dilations_0, groups = up_states_13_groups_0, pad = up_states_13_pad_0, pad_type = up_states_13_pad_type_0, strides = up_states_13_strides_0, weight = model_model_layers_6_mlp_up_proj_weight_palettized, x = input_93)[name = string("up_states_13")];
            tensor<fp16, [1, 8192, 1, 64]> gate_states_13 = silu(x = input_95)[name = string("gate_states_13")];
            tensor<fp16, [1, 8192, 1, 64]> input_97 = mul(x = gate_states_13, y = up_states_13)[name = string("input_97")];
            string hidden_states_55_pad_type_0 = const()[name = string("hidden_states_55_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_55_strides_0 = const()[name = string("hidden_states_55_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_55_pad_0 = const()[name = string("hidden_states_55_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_55_dilations_0 = const()[name = string("hidden_states_55_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_55_groups_0 = const()[name = string("hidden_states_55_groups_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 1, 64]> hidden_states_55 = conv(dilations = hidden_states_55_dilations_0, groups = hidden_states_55_groups_0, pad = hidden_states_55_pad_0, pad_type = hidden_states_55_pad_type_0, strides = hidden_states_55_strides_0, weight = model_model_layers_6_mlp_down_proj_weight_palettized, x = input_97)[name = string("hidden_states_55")];
            tensor<int32, [1]> var_1646_axes_0 = const()[name = string("op_1646_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 64]> var_1646 = squeeze(axes = var_1646_axes_0, x = hidden_states_55)[name = string("op_1646")];
            tensor<int32, [3]> var_1647 = const()[name = string("op_1647"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 64, 2048]> var_1648 = transpose(perm = var_1647, x = var_1646)[name = string("transpose_7")];
            tensor<fp16, [1, 64, 2048]> hidden_states_57_cast_fp16 = add(x = hidden_states_53_cast_fp16, y = var_1648)[name = string("hidden_states_57_cast_fp16")];
            tensor<int32, [1]> mean_29_axes_0 = const()[name = string("mean_29_axes_0"), val = tensor<int32, [1]>([-1])];
            bool mean_29_keep_dims_0 = const()[name = string("mean_29_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 64, 1]> mean_29_cast_fp16 = reduce_mean(axes = mean_29_axes_0, keep_dims = mean_29_keep_dims_0, x = hidden_states_57_cast_fp16)[name = string("mean_29_cast_fp16")];
            tensor<fp16, [1, 64, 2048]> input_99_cast_fp16 = sub(x = hidden_states_57_cast_fp16, y = mean_29_cast_fp16)[name = string("input_99_cast_fp16")];
            tensor<int32, [1]> var_1656_axes_0 = const()[name = string("op_1656_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2048]> model_model_layers_7_input_layernorm_weight_to_fp16 = const()[name = string("model_model_layers_7_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275542016)))];
            tensor<fp16, [1, 64, 2048]> var_1656_cast_fp16 = layer_norm(axes = var_1656_axes_0, epsilon = var_46_to_fp16, gamma = model_model_layers_7_input_layernorm_weight_to_fp16, x = input_99_cast_fp16)[name = string("op_1656_cast_fp16")];
            tensor<int32, [3]> var_1660 = const()[name = string("op_1660"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1662_axes_0 = const()[name = string("op_1662_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 64]> var_1661 = transpose(perm = var_1660, x = var_1656_cast_fp16)[name = string("transpose_6")];
            tensor<fp16, [1, 2048, 1, 64]> var_1662 = expand_dims(axes = var_1662_axes_0, x = var_1661)[name = string("op_1662")];
            string query_states_29_pad_type_0 = const()[name = string("query_states_29_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_states_29_strides_0 = const()[name = string("query_states_29_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_states_29_pad_0 = const()[name = string("query_states_29_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_states_29_dilations_0 = const()[name = string("query_states_29_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_states_29_groups_0 = const()[name = string("query_states_29_groups_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 1, 64]> query_states_29 = conv(dilations = query_states_29_dilations_0, groups = query_states_29_groups_0, pad = query_states_29_pad_0, pad_type = query_states_29_pad_type_0, strides = query_states_29_strides_0, weight = model_model_layers_7_self_attn_q_proj_weight_palettized, x = var_1662)[name = string("query_states_29")];
            string key_states_43_pad_type_0 = const()[name = string("key_states_43_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_states_43_strides_0 = const()[name = string("key_states_43_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_states_43_pad_0 = const()[name = string("key_states_43_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_states_43_dilations_0 = const()[name = string("key_states_43_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_states_43_groups_0 = const()[name = string("key_states_43_groups_0"), val = int32(1)];
            tensor<fp16, [1, 512, 1, 64]> key_states_43 = conv(dilations = key_states_43_dilations_0, groups = key_states_43_groups_0, pad = key_states_43_pad_0, pad_type = key_states_43_pad_type_0, strides = key_states_43_strides_0, weight = model_model_layers_7_self_attn_k_proj_weight_palettized, x = var_1662)[name = string("key_states_43")];
            string value_states_43_pad_type_0 = const()[name = string("value_states_43_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_states_43_strides_0 = const()[name = string("value_states_43_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_states_43_pad_0 = const()[name = string("value_states_43_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_states_43_dilations_0 = const()[name = string("value_states_43_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_states_43_groups_0 = const()[name = string("value_states_43_groups_0"), val = int32(1)];
            tensor<fp16, [1, 512, 1, 64]> value_states_43 = conv(dilations = value_states_43_dilations_0, groups = value_states_43_groups_0, pad = value_states_43_pad_0, pad_type = value_states_43_pad_type_0, strides = value_states_43_strides_0, weight = model_model_layers_7_self_attn_v_proj_weight_palettized, x = var_1662)[name = string("value_states_43")];
            tensor<int32, [4]> var_1682 = const()[name = string("op_1682"), val = tensor<int32, [4]>([1, 32, 64, 64])];
            tensor<fp16, [1, 32, 64, 64]> var_1683 = reshape(shape = var_1682, x = query_states_29)[name = string("op_1683")];
            tensor<int32, [4]> var_1684 = const()[name = string("op_1684"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_1686 = const()[name = string("op_1686"), val = tensor<int32, [4]>([1, 8, 64, 64])];
            tensor<fp16, [1, 8, 64, 64]> var_1687 = reshape(shape = var_1686, x = key_states_43)[name = string("op_1687")];
            tensor<int32, [4]> var_1688 = const()[name = string("op_1688"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_1690 = const()[name = string("op_1690"), val = tensor<int32, [4]>([1, 8, 64, 64])];
            tensor<fp16, [1, 8, 64, 64]> var_1691 = reshape(shape = var_1690, x = value_states_43)[name = string("op_1691")];
            tensor<int32, [4]> var_1692 = const()[name = string("op_1692"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> x1_29_begin_0 = const()[name = string("x1_29_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_29_end_0 = const()[name = string("x1_29_end_0"), val = tensor<int32, [4]>([1, 32, 64, 32])];
            tensor<bool, [4]> x1_29_end_mask_0 = const()[name = string("x1_29_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 64, 64]> x_197 = transpose(perm = var_1684, x = var_1683)[name = string("transpose_5")];
            tensor<fp16, [1, 32, 64, 32]> x1_29 = slice_by_index(begin = x1_29_begin_0, end = x1_29_end_0, end_mask = x1_29_end_mask_0, x = x_197)[name = string("x1_29")];
            tensor<int32, [4]> x2_29_begin_0 = const()[name = string("x2_29_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_29_end_0 = const()[name = string("x2_29_end_0"), val = tensor<int32, [4]>([1, 32, 64, 64])];
            tensor<bool, [4]> x2_29_end_mask_0 = const()[name = string("x2_29_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 64, 32]> x2_29 = slice_by_index(begin = x2_29_begin_0, end = x2_29_end_0, end_mask = x2_29_end_mask_0, x = x_197)[name = string("x2_29")];
            tensor<fp16, [1, 32, 64, 32]> var_1710 = mul(x = x1_29, y = cos_7)[name = string("op_1710")];
            tensor<fp16, [1, 32, 64, 32]> var_1711 = mul(x = x2_29, y = sin_7)[name = string("op_1711")];
            tensor<fp16, [1, 32, 64, 32]> var_1712 = sub(x = var_1710, y = var_1711)[name = string("op_1712")];
            tensor<fp16, [1, 32, 64, 32]> var_1713 = mul(x = x2_29, y = cos_7)[name = string("op_1713")];
            tensor<fp16, [1, 32, 64, 32]> var_1714 = mul(x = x1_29, y = sin_7)[name = string("op_1714")];
            tensor<fp16, [1, 32, 64, 32]> var_1715 = add(x = var_1713, y = var_1714)[name = string("op_1715")];
            bool rotated_29_interleave_0 = const()[name = string("rotated_29_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 32, 64, 64]> rotated_29 = concat(axis = var_44, interleave = rotated_29_interleave_0, values = (var_1712, var_1715))[name = string("rotated_29")];
            tensor<int32, [4]> x1_begin_0 = const()[name = string("x1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_end_0 = const()[name = string("x1_end_0"), val = tensor<int32, [4]>([1, 8, 64, 32])];
            tensor<bool, [4]> x1_end_mask_0 = const()[name = string("x1_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 64, 64]> x_201 = transpose(perm = var_1688, x = var_1687)[name = string("transpose_4")];
            tensor<fp16, [1, 8, 64, 32]> x1 = slice_by_index(begin = x1_begin_0, end = x1_end_0, end_mask = x1_end_mask_0, x = x_201)[name = string("x1")];
            tensor<int32, [4]> x2_begin_0 = const()[name = string("x2_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_end_0 = const()[name = string("x2_end_0"), val = tensor<int32, [4]>([1, 8, 64, 64])];
            tensor<bool, [4]> x2_end_mask_0 = const()[name = string("x2_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 64, 32]> x2 = slice_by_index(begin = x2_begin_0, end = x2_end_0, end_mask = x2_end_mask_0, x = x_201)[name = string("x2")];
            tensor<fp16, [1, 8, 64, 32]> var_1731 = mul(x = x1, y = cos_7)[name = string("op_1731")];
            tensor<fp16, [1, 8, 64, 32]> var_1732 = mul(x = x2, y = sin_7)[name = string("op_1732")];
            tensor<fp16, [1, 8, 64, 32]> var_1733 = sub(x = var_1731, y = var_1732)[name = string("op_1733")];
            tensor<fp16, [1, 8, 64, 32]> var_1734 = mul(x = x2, y = cos_7)[name = string("op_1734")];
            tensor<fp16, [1, 8, 64, 32]> var_1735 = mul(x = x1, y = sin_7)[name = string("op_1735")];
            tensor<fp16, [1, 8, 64, 32]> var_1736 = add(x = var_1734, y = var_1735)[name = string("op_1736")];
            bool rotated_interleave_0 = const()[name = string("rotated_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 8, 64, 64]> rotated = concat(axis = var_44, interleave = rotated_interleave_0, values = (var_1733, var_1736))[name = string("rotated")];
            tensor<int32, [1]> expand_dims_84 = const()[name = string("expand_dims_84"), val = tensor<int32, [1]>([7])];
            tensor<int32, [1]> expand_dims_85 = const()[name = string("expand_dims_85"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_87 = const()[name = string("expand_dims_87"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_88 = const()[name = string("expand_dims_88"), val = tensor<int32, [1]>([8])];
            int32 concat_128_axis_0 = const()[name = string("concat_128_axis_0"), val = int32(0)];
            bool concat_128_interleave_0 = const()[name = string("concat_128_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_128 = concat(axis = concat_128_axis_0, interleave = concat_128_interleave_0, values = (expand_dims_84, expand_dims_85, current_pos, expand_dims_87))[name = string("concat_128")];
            tensor<int32, [1]> concat_129_values1_0 = const()[name = string("concat_129_values1_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> concat_129_values3_0 = const()[name = string("concat_129_values3_0"), val = tensor<int32, [1]>([0])];
            int32 concat_129_axis_0 = const()[name = string("concat_129_axis_0"), val = int32(0)];
            bool concat_129_interleave_0 = const()[name = string("concat_129_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_129 = concat(axis = concat_129_axis_0, interleave = concat_129_interleave_0, values = (expand_dims_88, concat_129_values1_0, var_366, concat_129_values3_0))[name = string("concat_129")];
            tensor<int32, [4]> model_model_kv_cache_0_internal_tensor_assign_15_stride_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_15_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_15_begin_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_15_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_15_end_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_15_end_mask_0"), val = tensor<bool, [4]>([false, true, false, true])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_15_squeeze_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_15_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<fp16, [32, 8, 2048, 64]> model_model_kv_cache_0_internal_tensor_assign_15_cast_fp16 = slice_update(begin = concat_128, begin_mask = model_model_kv_cache_0_internal_tensor_assign_15_begin_mask_0, end = concat_129, end_mask = model_model_kv_cache_0_internal_tensor_assign_15_end_mask_0, squeeze_mask = model_model_kv_cache_0_internal_tensor_assign_15_squeeze_mask_0, stride = model_model_kv_cache_0_internal_tensor_assign_15_stride_0, update = rotated, x = coreml_update_state_29)[name = string("model_model_kv_cache_0_internal_tensor_assign_15_cast_fp16")];
            write_state(data = model_model_kv_cache_0_internal_tensor_assign_15_cast_fp16, input = model_model_kv_cache_0)[name = string("coreml_update_state_30_write_state")];
            tensor<fp16, [32, 8, 2048, 64]> coreml_update_state_30 = read_state(input = model_model_kv_cache_0)[name = string("coreml_update_state_30")];
            tensor<int32, [1]> expand_dims_90 = const()[name = string("expand_dims_90"), val = tensor<int32, [1]>([23])];
            tensor<int32, [1]> expand_dims_91 = const()[name = string("expand_dims_91"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_93 = const()[name = string("expand_dims_93"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> expand_dims_94 = const()[name = string("expand_dims_94"), val = tensor<int32, [1]>([24])];
            int32 concat_132_axis_0 = const()[name = string("concat_132_axis_0"), val = int32(0)];
            bool concat_132_interleave_0 = const()[name = string("concat_132_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_132 = concat(axis = concat_132_axis_0, interleave = concat_132_interleave_0, values = (expand_dims_90, expand_dims_91, current_pos, expand_dims_93))[name = string("concat_132")];
            tensor<int32, [1]> concat_133_values1_0 = const()[name = string("concat_133_values1_0"), val = tensor<int32, [1]>([0])];
            tensor<int32, [1]> concat_133_values3_0 = const()[name = string("concat_133_values3_0"), val = tensor<int32, [1]>([0])];
            int32 concat_133_axis_0 = const()[name = string("concat_133_axis_0"), val = int32(0)];
            bool concat_133_interleave_0 = const()[name = string("concat_133_interleave_0"), val = bool(false)];
            tensor<int32, [4]> concat_133 = concat(axis = concat_133_axis_0, interleave = concat_133_interleave_0, values = (expand_dims_94, concat_133_values1_0, var_366, concat_133_values3_0))[name = string("concat_133")];
            tensor<int32, [4]> model_model_kv_cache_0_internal_tensor_assign_16_stride_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_16_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_16_begin_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_16_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_16_end_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_16_end_mask_0"), val = tensor<bool, [4]>([false, true, false, true])];
            tensor<bool, [4]> model_model_kv_cache_0_internal_tensor_assign_16_squeeze_mask_0 = const()[name = string("model_model_kv_cache_0_internal_tensor_assign_16_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])];
            tensor<fp16, [1, 8, 64, 64]> value_states_45 = transpose(perm = var_1692, x = var_1691)[name = string("transpose_3")];
            tensor<fp16, [32, 8, 2048, 64]> model_model_kv_cache_0_internal_tensor_assign_16_cast_fp16 = slice_update(begin = concat_132, begin_mask = model_model_kv_cache_0_internal_tensor_assign_16_begin_mask_0, end = concat_133, end_mask = model_model_kv_cache_0_internal_tensor_assign_16_end_mask_0, squeeze_mask = model_model_kv_cache_0_internal_tensor_assign_16_squeeze_mask_0, stride = model_model_kv_cache_0_internal_tensor_assign_16_stride_0, update = value_states_45, x = coreml_update_state_30)[name = string("model_model_kv_cache_0_internal_tensor_assign_16_cast_fp16")];
            write_state(data = model_model_kv_cache_0_internal_tensor_assign_16_cast_fp16, input = model_model_kv_cache_0)[name = string("coreml_update_state_31_write_state")];
            tensor<fp16, [32, 8, 2048, 64]> coreml_update_state_31 = read_state(input = model_model_kv_cache_0)[name = string("coreml_update_state_31")];
            tensor<int32, [4]> var_1759_begin_0 = const()[name = string("op_1759_begin_0"), val = tensor<int32, [4]>([7, 0, 0, 0])];
            tensor<int32, [4]> var_1759_end_0 = const()[name = string("op_1759_end_0"), val = tensor<int32, [4]>([8, 8, 2048, 64])];
            tensor<bool, [4]> var_1759_end_mask_0 = const()[name = string("op_1759_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<fp16, [1, 8, 2048, 64]> var_1759_cast_fp16 = slice_by_index(begin = var_1759_begin_0, end = var_1759_end_0, end_mask = var_1759_end_mask_0, x = coreml_update_state_31)[name = string("op_1759_cast_fp16")];
            tensor<int32, [1]> K_layer_cache_axes_0 = const()[name = string("K_layer_cache_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<fp16, [8, 2048, 64]> K_layer_cache_cast_fp16 = squeeze(axes = K_layer_cache_axes_0, x = var_1759_cast_fp16)[name = string("K_layer_cache_cast_fp16")];
            tensor<int32, [4]> var_1761_begin_0 = const()[name = string("op_1761_begin_0"), val = tensor<int32, [4]>([23, 0, 0, 0])];
            tensor<int32, [4]> var_1761_end_0 = const()[name = string("op_1761_end_0"), val = tensor<int32, [4]>([24, 8, 2048, 64])];
            tensor<bool, [4]> var_1761_end_mask_0 = const()[name = string("op_1761_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<fp16, [1, 8, 2048, 64]> var_1761_cast_fp16 = slice_by_index(begin = var_1761_begin_0, end = var_1761_end_0, end_mask = var_1761_end_mask_0, x = coreml_update_state_31)[name = string("op_1761_cast_fp16")];
            tensor<int32, [1]> V_layer_cache_axes_0 = const()[name = string("V_layer_cache_axes_0"), val = tensor<int32, [1]>([0])];
            tensor<fp16, [8, 2048, 64]> V_layer_cache_cast_fp16 = squeeze(axes = V_layer_cache_axes_0, x = var_1761_cast_fp16)[name = string("V_layer_cache_cast_fp16")];
            tensor<int32, [1]> x_207_axes_0 = const()[name = string("x_207_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [8, 1, 2048, 64]> x_207_cast_fp16 = expand_dims(axes = x_207_axes_0, x = K_layer_cache_cast_fp16)[name = string("x_207_cast_fp16")];
            tensor<int32, [4]> var_1770 = const()[name = string("op_1770"), val = tensor<int32, [4]>([1, 4, 1, 1])];
            tensor<fp16, [8, 4, 2048, 64]> x_209_cast_fp16 = tile(reps = var_1770, x = x_207_cast_fp16)[name = string("x_209_cast_fp16")];
            tensor<int32, [4]> var_1774 = const()[name = string("op_1774"), val = tensor<int32, [4]>([1, -1, 2048, 64])];
            tensor<fp16, [1, 32, 2048, 64]> var_1775_cast_fp16 = reshape(shape = var_1774, x = x_209_cast_fp16)[name = string("op_1775_cast_fp16")];
            tensor<int32, [1]> x_213_axes_0 = const()[name = string("x_213_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [8, 1, 2048, 64]> x_213_cast_fp16 = expand_dims(axes = x_213_axes_0, x = V_layer_cache_cast_fp16)[name = string("x_213_cast_fp16")];
            tensor<int32, [4]> var_1777 = const()[name = string("op_1777"), val = tensor<int32, [4]>([1, 4, 1, 1])];
            tensor<fp16, [8, 4, 2048, 64]> x_215_cast_fp16 = tile(reps = var_1777, x = x_213_cast_fp16)[name = string("x_215_cast_fp16")];
            bool var_1784_transpose_x_0 = const()[name = string("op_1784_transpose_x_0"), val = bool(false)];
            bool var_1784_transpose_y_0 = const()[name = string("op_1784_transpose_y_0"), val = bool(true)];
            tensor<fp16, [1, 32, 64, 2048]> var_1784_cast_fp16 = matmul(transpose_x = var_1784_transpose_x_0, transpose_y = var_1784_transpose_y_0, x = rotated_29, y = var_1775_cast_fp16)[name = string("op_1784_cast_fp16")];
            fp16 var_1785_to_fp16 = const()[name = string("op_1785_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 32, 64, 2048]> attn_weights_cast_fp16 = mul(x = var_1784_cast_fp16, y = var_1785_to_fp16)[name = string("attn_weights_cast_fp16")];
            tensor<fp16, [1, 32, 64, 2048]> x_217_cast_fp16 = add(x = attn_weights_cast_fp16, y = causal_mask)[name = string("x_217_cast_fp16")];
            tensor<int32, [1]> reduce_max_7_axes_0 = const()[name = string("reduce_max_7_axes_0"), val = tensor<int32, [1]>([-1])];
            bool reduce_max_7_keep_dims_0 = const()[name = string("reduce_max_7_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 64, 1]> reduce_max_7_cast_fp16 = reduce_max(axes = reduce_max_7_axes_0, keep_dims = reduce_max_7_keep_dims_0, x = x_217_cast_fp16)[name = string("reduce_max_7_cast_fp16")];
            tensor<fp16, [1, 32, 64, 2048]> x_219_cast_fp16 = sub(x = x_217_cast_fp16, y = reduce_max_7_cast_fp16)[name = string("x_219_cast_fp16")];
            tensor<fp16, [1, 32, 64, 2048]> exp_x_cast_fp16 = exp(x = x_219_cast_fp16)[name = string("exp_x_cast_fp16")];
            tensor<int32, [1]> var_1796_axes_0 = const()[name = string("op_1796_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1796_keep_dims_0 = const()[name = string("op_1796_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 32, 64, 1]> var_1796_cast_fp16 = reduce_sum(axes = var_1796_axes_0, keep_dims = var_1796_keep_dims_0, x = exp_x_cast_fp16)[name = string("op_1796_cast_fp16")];
            tensor<fp16, [1, 32, 64, 2048]> var_1797_cast_fp16 = real_div(x = exp_x_cast_fp16, y = var_1796_cast_fp16)[name = string("op_1797_cast_fp16")];
            tensor<int32, [3]> concat_138 = const()[name = string("concat_138"), val = tensor<int32, [3]>([32, 64, 2048])];
            tensor<fp16, [32, 64, 2048]> reshape_21_cast_fp16 = reshape(shape = concat_138, x = var_1797_cast_fp16)[name = string("reshape_21_cast_fp16")];
            tensor<int32, [3]> concat_139 = const()[name = string("concat_139"), val = tensor<int32, [3]>([32, 2048, 64])];
            tensor<fp16, [32, 2048, 64]> reshape_22_cast_fp16 = reshape(shape = concat_139, x = x_215_cast_fp16)[name = string("reshape_22_cast_fp16")];
            bool matmul_7_transpose_x_0 = const()[name = string("matmul_7_transpose_x_0"), val = bool(false)];
            bool matmul_7_transpose_y_0 = const()[name = string("matmul_7_transpose_y_0"), val = bool(false)];
            tensor<fp16, [32, 64, 64]> matmul_7_cast_fp16 = matmul(transpose_x = matmul_7_transpose_x_0, transpose_y = matmul_7_transpose_y_0, x = reshape_21_cast_fp16, y = reshape_22_cast_fp16)[name = string("matmul_7_cast_fp16")];
            tensor<int32, [4]> concat_143 = const()[name = string("concat_143"), val = tensor<int32, [4]>([1, 32, 64, 64])];
            tensor<fp16, [1, 32, 64, 64]> reshape_23_cast_fp16 = reshape(shape = concat_143, x = matmul_7_cast_fp16)[name = string("reshape_23_cast_fp16")];
            tensor<int32, [4]> var_1800_perm_0 = const()[name = string("op_1800_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1802 = const()[name = string("op_1802"), val = tensor<int32, [3]>([1, 64, 2048])];
            tensor<fp16, [1, 64, 32, 64]> var_1800_cast_fp16 = transpose(perm = var_1800_perm_0, x = reshape_23_cast_fp16)[name = string("transpose_2")];
            tensor<fp16, [1, 64, 2048]> input_103_cast_fp16 = reshape(shape = var_1802, x = var_1800_cast_fp16)[name = string("input_103_cast_fp16")];
            tensor<fp16, [2048, 2048]> model_model_layers_7_self_attn_o_proj_weight_promoted_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint4, [2048, 2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(275546176))), lut = tensor<fp16, [256, 1, 16, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277643392))))[name = string("model_model_layers_7_self_attn_o_proj_weight_promoted_to_fp16_palettized")];
            tensor<fp16, [1, 64, 2048]> linear_7_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_7_self_attn_o_proj_weight_promoted_to_fp16_palettized, x = input_103_cast_fp16)[name = string("linear_7_cast_fp16")];
            tensor<fp16, [1, 64, 2048]> hidden_states_61_cast_fp16 = add(x = hidden_states_57_cast_fp16, y = linear_7_cast_fp16)[name = string("hidden_states_61_cast_fp16")];
            tensor<int32, [1]> mean_axes_0 = const()[name = string("mean_axes_0"), val = tensor<int32, [1]>([-1])];
            bool mean_keep_dims_0 = const()[name = string("mean_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 64, 1]> mean_cast_fp16 = reduce_mean(axes = mean_axes_0, keep_dims = mean_keep_dims_0, x = hidden_states_61_cast_fp16)[name = string("mean_cast_fp16")];
            tensor<fp16, [1, 64, 2048]> input_105_cast_fp16 = sub(x = hidden_states_61_cast_fp16, y = mean_cast_fp16)[name = string("input_105_cast_fp16")];
            tensor<int32, [1]> var_1813_axes_0 = const()[name = string("op_1813_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [2048]> model_model_layers_7_post_attention_layernorm_weight_to_fp16 = const()[name = string("model_model_layers_7_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277651648)))];
            tensor<fp16, [1, 64, 2048]> var_1813_cast_fp16 = layer_norm(axes = var_1813_axes_0, epsilon = var_46_to_fp16, gamma = model_model_layers_7_post_attention_layernorm_weight_to_fp16, x = input_105_cast_fp16)[name = string("op_1813_cast_fp16")];
            tensor<int32, [3]> var_1820 = const()[name = string("op_1820"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> input_107_axes_0 = const()[name = string("input_107_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 64]> var_1821 = transpose(perm = var_1820, x = var_1813_cast_fp16)[name = string("transpose_1")];
            tensor<fp16, [1, 2048, 1, 64]> input_107 = expand_dims(axes = input_107_axes_0, x = var_1821)[name = string("input_107")];
            string input_109_pad_type_0 = const()[name = string("input_109_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_109_strides_0 = const()[name = string("input_109_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_109_pad_0 = const()[name = string("input_109_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_109_dilations_0 = const()[name = string("input_109_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_109_groups_0 = const()[name = string("input_109_groups_0"), val = int32(1)];
            tensor<fp16, [1, 8192, 1, 64]> input_109 = conv(dilations = input_109_dilations_0, groups = input_109_groups_0, pad = input_109_pad_0, pad_type = input_109_pad_type_0, strides = input_109_strides_0, weight = model_model_layers_7_mlp_gate_proj_weight_palettized, x = input_107)[name = string("input_109")];
            string up_states_pad_type_0 = const()[name = string("up_states_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> up_states_strides_0 = const()[name = string("up_states_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> up_states_pad_0 = const()[name = string("up_states_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> up_states_dilations_0 = const()[name = string("up_states_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 up_states_groups_0 = const()[name = string("up_states_groups_0"), val = int32(1)];
            tensor<fp16, [1, 8192, 1, 64]> up_states = conv(dilations = up_states_dilations_0, groups = up_states_groups_0, pad = up_states_pad_0, pad_type = up_states_pad_type_0, strides = up_states_strides_0, weight = model_model_layers_7_mlp_up_proj_weight_palettized, x = input_107)[name = string("up_states")];
            tensor<fp16, [1, 8192, 1, 64]> gate_states = silu(x = input_109)[name = string("gate_states")];
            tensor<fp16, [1, 8192, 1, 64]> input = mul(x = gate_states, y = up_states)[name = string("input")];
            string hidden_states_pad_type_0 = const()[name = string("hidden_states_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_strides_0 = const()[name = string("hidden_states_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_pad_0 = const()[name = string("hidden_states_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_dilations_0 = const()[name = string("hidden_states_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_groups_0 = const()[name = string("hidden_states_groups_0"), val = int32(1)];
            tensor<fp16, [1, 2048, 1, 64]> hidden_states_1 = conv(dilations = hidden_states_dilations_0, groups = hidden_states_groups_0, pad = hidden_states_pad_0, pad_type = hidden_states_pad_type_0, strides = hidden_states_strides_0, weight = model_model_layers_7_mlp_down_proj_weight_palettized, x = input)[name = string("hidden_states")];
            tensor<int32, [1]> var_1843_axes_0 = const()[name = string("op_1843_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 2048, 64]> var_1843 = squeeze(axes = var_1843_axes_0, x = hidden_states_1)[name = string("op_1843")];
            tensor<int32, [3]> var_1844 = const()[name = string("op_1844"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 64, 2048]> var_1845 = transpose(perm = var_1844, x = var_1843)[name = string("transpose_0")];
            tensor<fp16, [1, 64, 2048]> output_hidden_states = add(x = hidden_states_61_cast_fp16, y = var_1845)[name = string("op_1846_cast_fp16")];
        } -> (output_hidden_states);
}