program(1.3) [buildInfo = dict({{"coremlc-component-MIL", "3400.43.1"}, {"coremlc-version", "3400.58.2"}, {"coremltools-component-torch", "2.4.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] { func main(tensor logmel_data) { string var_32_pad_type_0 = const()[name = string("op_32_pad_type_0"), val = string("custom")]; tensor var_32_pad_0 = const()[name = string("op_32_pad_0"), val = tensor([1, 1])]; tensor var_32_strides_0 = const()[name = string("op_32_strides_0"), val = tensor([1])]; tensor var_32_dilations_0 = const()[name = string("op_32_dilations_0"), val = tensor([1])]; int32 var_32_groups_0 = const()[name = string("op_32_groups_0"), val = int32(1)]; tensor weight_3_to_fp16 = const()[name = string("weight_3_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; tensor bias_3_to_fp16 = const()[name = string("bias_3_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245888)))]; tensor var_32_cast_fp16 = conv(bias = bias_3_to_fp16, dilations = var_32_dilations_0, groups = var_32_groups_0, pad = var_32_pad_0, pad_type = var_32_pad_type_0, strides = var_32_strides_0, weight = weight_3_to_fp16, x = logmel_data)[name = string("op_32_cast_fp16")]; string input_1_mode_0 = const()[name = string("input_1_mode_0"), val = string("EXACT")]; tensor input_1_cast_fp16 = gelu(mode = input_1_mode_0, x = var_32_cast_fp16)[name = string("input_1_cast_fp16")]; string var_50_pad_type_0 = const()[name = string("op_50_pad_type_0"), val = string("custom")]; tensor var_50_pad_0 = const()[name = string("op_50_pad_0"), val = tensor([1, 1])]; tensor var_50_strides_0 = const()[name = string("op_50_strides_0"), val = tensor([2])]; tensor var_50_dilations_0 = const()[name = string("op_50_dilations_0"), val = tensor([1])]; int32 var_50_groups_0 = const()[name = string("op_50_groups_0"), val = int32(1)]; tensor weight_7_to_fp16 = const()[name = string("weight_7_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246976)))]; tensor bias_7_to_fp16 = const()[name = string("bias_7_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1819904)))]; tensor var_50_cast_fp16 = conv(bias = bias_7_to_fp16, dilations = var_50_dilations_0, groups = var_50_groups_0, pad = var_50_pad_0, pad_type = var_50_pad_type_0, strides = var_50_strides_0, weight = weight_7_to_fp16, x = input_1_cast_fp16)[name = string("op_50_cast_fp16")]; string x_3_mode_0 = const()[name = string("x_3_mode_0"), val = string("EXACT")]; tensor x_3_cast_fp16 = gelu(mode = x_3_mode_0, x = var_50_cast_fp16)[name = string("x_3_cast_fp16")]; tensor var_56 = const()[name = string("op_56"), val = tensor([0, 2, 1])]; tensor positional_embedding_to_fp16 = const()[name = string("positional_embedding_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1820992)))]; tensor x_5_cast_fp16 = transpose(perm = var_56, x = x_3_cast_fp16)[name = string("transpose_60")]; tensor var_59_cast_fp16 = add(x = x_5_cast_fp16, y = positional_embedding_to_fp16)[name = string("op_59_cast_fp16")]; int32 var_72 = const()[name = string("op_72"), val = int32(-1)]; tensor var_88_axes_0 = const()[name = string("op_88_axes_0"), val = tensor([-1])]; tensor blocks_0_attn_ln_weight_to_fp16 = const()[name = string("blocks_0_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3357056)))]; tensor blocks_0_attn_ln_bias_to_fp16 = const()[name = string("blocks_0_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3358144)))]; fp16 var_78_to_fp16 = const()[name = string("op_78_to_fp16"), val = fp16(0x1.5p-17)]; tensor var_88_cast_fp16 = layer_norm(axes = var_88_axes_0, beta = blocks_0_attn_ln_bias_to_fp16, epsilon = var_78_to_fp16, gamma = blocks_0_attn_ln_weight_to_fp16, x = var_59_cast_fp16)[name = string("op_88_cast_fp16")]; tensor var_99_to_fp16 = const()[name = string("op_99_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3359232)))]; tensor var_100_to_fp16 = const()[name = string("op_100_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3883584)))]; tensor linear_0_cast_fp16 = linear(bias = var_100_to_fp16, weight = var_99_to_fp16, x = var_88_cast_fp16)[name = string("linear_0_cast_fp16")]; tensor var_103_to_fp16 = const()[name = string("op_103_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3884672)))]; tensor linear_1_bias_0_to_fp16 = const()[name = string("linear_1_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4409024)))]; tensor linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_103_to_fp16, x = var_88_cast_fp16)[name = string("linear_1_cast_fp16")]; tensor var_107_to_fp16 = const()[name = string("op_107_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4410112)))]; tensor var_108_to_fp16 = const()[name = string("op_108_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4934464)))]; tensor linear_2_cast_fp16 = linear(bias = var_108_to_fp16, weight = var_107_to_fp16, x = var_88_cast_fp16)[name = string("linear_2_cast_fp16")]; tensor var_116 = const()[name = string("op_116"), val = tensor([1, 1500, 8, -1])]; tensor var_117_cast_fp16 = reshape(shape = var_116, x = linear_0_cast_fp16)[name = string("op_117_cast_fp16")]; tensor const_42_to_fp16 = const()[name = string("const_42_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; tensor q_3_cast_fp16 = mul(x = var_117_cast_fp16, y = const_42_to_fp16)[name = string("q_3_cast_fp16")]; tensor var_123 = const()[name = string("op_123"), val = tensor([1, 1500, 8, -1])]; tensor var_124_cast_fp16 = reshape(shape = var_123, x = linear_1_cast_fp16)[name = string("op_124_cast_fp16")]; tensor const_43_to_fp16 = const()[name = string("const_43_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; tensor k_3_cast_fp16 = mul(x = var_124_cast_fp16, y = const_43_to_fp16)[name = string("k_3_cast_fp16")]; tensor var_130 = const()[name = string("op_130"), val = tensor([1, 1500, 8, -1])]; tensor var_131_cast_fp16 = reshape(shape = var_130, x = linear_2_cast_fp16)[name = string("op_131_cast_fp16")]; tensor var_132 = const()[name = string("op_132"), val = tensor([0, 2, 1, 3])]; bool qk_1_transpose_x_0 = const()[name = string("qk_1_transpose_x_0"), val = bool(false)]; bool qk_1_transpose_y_0 = const()[name = string("qk_1_transpose_y_0"), val = bool(false)]; tensor transpose_24_perm_0 = const()[name = string("transpose_24_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_25_perm_0 = const()[name = string("transpose_25_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_25 = transpose(perm = transpose_25_perm_0, x = k_3_cast_fp16)[name = string("transpose_57")]; tensor transpose_24 = transpose(perm = transpose_24_perm_0, x = q_3_cast_fp16)[name = string("transpose_58")]; tensor qk_1_cast_fp16 = matmul(transpose_x = qk_1_transpose_x_0, transpose_y = qk_1_transpose_y_0, x = transpose_24, y = transpose_25)[name = string("qk_1_cast_fp16")]; tensor var_136_cast_fp16 = softmax(axis = var_72, x = qk_1_cast_fp16)[name = string("op_136_cast_fp16")]; bool var_138_transpose_x_0 = const()[name = string("op_138_transpose_x_0"), val = bool(false)]; bool var_138_transpose_y_0 = const()[name = string("op_138_transpose_y_0"), val = bool(false)]; tensor v_3_cast_fp16 = transpose(perm = var_132, x = var_131_cast_fp16)[name = string("transpose_59")]; tensor var_138_cast_fp16 = matmul(transpose_x = var_138_transpose_x_0, transpose_y = var_138_transpose_y_0, x = var_136_cast_fp16, y = v_3_cast_fp16)[name = string("op_138_cast_fp16")]; tensor var_139 = const()[name = string("op_139"), val = tensor([0, 2, 1, 3])]; tensor concat_0 = const()[name = string("concat_0"), val = tensor([1, 1500, 512])]; tensor var_140_cast_fp16 = transpose(perm = var_139, x = var_138_cast_fp16)[name = string("transpose_56")]; tensor x_11_cast_fp16 = reshape(shape = concat_0, x = var_140_cast_fp16)[name = string("x_11_cast_fp16")]; tensor var_144_to_fp16 = const()[name = string("op_144_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4935552)))]; tensor var_145_to_fp16 = const()[name = string("op_145_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5459904)))]; tensor linear_3_cast_fp16 = linear(bias = var_145_to_fp16, weight = var_144_to_fp16, x = x_11_cast_fp16)[name = string("linear_3_cast_fp16")]; tensor x_13_cast_fp16 = add(x = var_59_cast_fp16, y = linear_3_cast_fp16)[name = string("x_13_cast_fp16")]; tensor var_152_axes_0 = const()[name = string("op_152_axes_0"), val = tensor([-1])]; tensor blocks_0_mlp_ln_weight_to_fp16 = const()[name = string("blocks_0_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5460992)))]; tensor blocks_0_mlp_ln_bias_to_fp16 = const()[name = string("blocks_0_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5462080)))]; tensor var_152_cast_fp16 = layer_norm(axes = var_152_axes_0, beta = blocks_0_mlp_ln_bias_to_fp16, epsilon = var_78_to_fp16, gamma = blocks_0_mlp_ln_weight_to_fp16, x = x_13_cast_fp16)[name = string("op_152_cast_fp16")]; tensor var_161_to_fp16 = const()[name = string("op_161_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5463168)))]; tensor var_162_to_fp16 = const()[name = string("op_162_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7560384)))]; tensor linear_4_cast_fp16 = linear(bias = var_162_to_fp16, weight = var_161_to_fp16, x = var_152_cast_fp16)[name = string("linear_4_cast_fp16")]; string x_17_mode_0 = const()[name = string("x_17_mode_0"), val = string("EXACT")]; tensor x_17_cast_fp16 = gelu(mode = x_17_mode_0, x = linear_4_cast_fp16)[name = string("x_17_cast_fp16")]; tensor var_167_to_fp16 = const()[name = string("op_167_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7564544)))]; tensor var_168_to_fp16 = const()[name = string("op_168_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9661760)))]; tensor linear_5_cast_fp16 = linear(bias = var_168_to_fp16, weight = var_167_to_fp16, x = x_17_cast_fp16)[name = string("linear_5_cast_fp16")]; tensor x_19_cast_fp16 = add(x = x_13_cast_fp16, y = linear_5_cast_fp16)[name = string("x_19_cast_fp16")]; int32 var_178 = const()[name = string("op_178"), val = int32(-1)]; tensor var_194_axes_0 = const()[name = string("op_194_axes_0"), val = tensor([-1])]; tensor blocks_1_attn_ln_weight_to_fp16 = const()[name = string("blocks_1_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9662848)))]; tensor blocks_1_attn_ln_bias_to_fp16 = const()[name = string("blocks_1_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9663936)))]; fp16 var_184_to_fp16 = const()[name = string("op_184_to_fp16"), val = fp16(0x1.5p-17)]; tensor var_194_cast_fp16 = layer_norm(axes = var_194_axes_0, beta = blocks_1_attn_ln_bias_to_fp16, epsilon = var_184_to_fp16, gamma = blocks_1_attn_ln_weight_to_fp16, x = x_19_cast_fp16)[name = string("op_194_cast_fp16")]; tensor var_205_to_fp16 = const()[name = string("op_205_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9665024)))]; tensor var_206_to_fp16 = const()[name = string("op_206_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10189376)))]; tensor linear_6_cast_fp16 = linear(bias = var_206_to_fp16, weight = var_205_to_fp16, x = var_194_cast_fp16)[name = string("linear_6_cast_fp16")]; tensor var_209_to_fp16 = const()[name = string("op_209_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10190464)))]; tensor linear_7_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_209_to_fp16, x = var_194_cast_fp16)[name = string("linear_7_cast_fp16")]; tensor var_213_to_fp16 = const()[name = string("op_213_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10714816)))]; tensor var_214_to_fp16 = const()[name = string("op_214_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11239168)))]; tensor linear_8_cast_fp16 = linear(bias = var_214_to_fp16, weight = var_213_to_fp16, x = var_194_cast_fp16)[name = string("linear_8_cast_fp16")]; tensor var_222 = const()[name = string("op_222"), val = tensor([1, 1500, 8, -1])]; tensor var_223_cast_fp16 = reshape(shape = var_222, x = linear_6_cast_fp16)[name = string("op_223_cast_fp16")]; tensor const_44_to_fp16 = const()[name = string("const_44_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; tensor q_7_cast_fp16 = mul(x = var_223_cast_fp16, y = const_44_to_fp16)[name = string("q_7_cast_fp16")]; tensor var_229 = const()[name = string("op_229"), val = tensor([1, 1500, 8, -1])]; tensor var_230_cast_fp16 = reshape(shape = var_229, x = linear_7_cast_fp16)[name = string("op_230_cast_fp16")]; tensor const_45_to_fp16 = const()[name = string("const_45_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; tensor k_7_cast_fp16 = mul(x = var_230_cast_fp16, y = const_45_to_fp16)[name = string("k_7_cast_fp16")]; tensor var_236 = const()[name = string("op_236"), val = tensor([1, 1500, 8, -1])]; tensor var_237_cast_fp16 = reshape(shape = var_236, x = linear_8_cast_fp16)[name = string("op_237_cast_fp16")]; tensor var_238 = const()[name = string("op_238"), val = tensor([0, 2, 1, 3])]; bool qk_3_transpose_x_0 = const()[name = string("qk_3_transpose_x_0"), val = bool(false)]; bool qk_3_transpose_y_0 = const()[name = string("qk_3_transpose_y_0"), val = bool(false)]; tensor transpose_26_perm_0 = const()[name = string("transpose_26_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_27_perm_0 = const()[name = string("transpose_27_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_27 = transpose(perm = transpose_27_perm_0, x = k_7_cast_fp16)[name = string("transpose_53")]; tensor transpose_26 = transpose(perm = transpose_26_perm_0, x = q_7_cast_fp16)[name = string("transpose_54")]; tensor qk_3_cast_fp16 = matmul(transpose_x = qk_3_transpose_x_0, transpose_y = qk_3_transpose_y_0, x = transpose_26, y = transpose_27)[name = string("qk_3_cast_fp16")]; tensor var_242_cast_fp16 = softmax(axis = var_178, x = qk_3_cast_fp16)[name = string("op_242_cast_fp16")]; bool var_244_transpose_x_0 = const()[name = string("op_244_transpose_x_0"), val = bool(false)]; bool var_244_transpose_y_0 = const()[name = string("op_244_transpose_y_0"), val = bool(false)]; tensor v_7_cast_fp16 = transpose(perm = var_238, x = var_237_cast_fp16)[name = string("transpose_55")]; tensor var_244_cast_fp16 = matmul(transpose_x = var_244_transpose_x_0, transpose_y = var_244_transpose_y_0, x = var_242_cast_fp16, y = v_7_cast_fp16)[name = string("op_244_cast_fp16")]; tensor var_245 = const()[name = string("op_245"), val = tensor([0, 2, 1, 3])]; tensor concat_1 = const()[name = string("concat_1"), val = tensor([1, 1500, 512])]; tensor var_246_cast_fp16 = transpose(perm = var_245, x = var_244_cast_fp16)[name = string("transpose_52")]; tensor x_23_cast_fp16 = reshape(shape = concat_1, x = var_246_cast_fp16)[name = string("x_23_cast_fp16")]; tensor var_250_to_fp16 = const()[name = string("op_250_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11240256)))]; tensor var_251_to_fp16 = const()[name = string("op_251_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11764608)))]; tensor linear_9_cast_fp16 = linear(bias = var_251_to_fp16, weight = var_250_to_fp16, x = x_23_cast_fp16)[name = string("linear_9_cast_fp16")]; tensor x_25_cast_fp16 = add(x = x_19_cast_fp16, y = linear_9_cast_fp16)[name = string("x_25_cast_fp16")]; tensor var_258_axes_0 = const()[name = string("op_258_axes_0"), val = tensor([-1])]; tensor blocks_1_mlp_ln_weight_to_fp16 = const()[name = string("blocks_1_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11765696)))]; tensor blocks_1_mlp_ln_bias_to_fp16 = const()[name = string("blocks_1_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11766784)))]; tensor var_258_cast_fp16 = layer_norm(axes = var_258_axes_0, beta = blocks_1_mlp_ln_bias_to_fp16, epsilon = var_184_to_fp16, gamma = blocks_1_mlp_ln_weight_to_fp16, x = x_25_cast_fp16)[name = string("op_258_cast_fp16")]; tensor var_267_to_fp16 = const()[name = string("op_267_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11767872)))]; tensor var_268_to_fp16 = const()[name = string("op_268_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13865088)))]; tensor linear_10_cast_fp16 = linear(bias = var_268_to_fp16, weight = var_267_to_fp16, x = var_258_cast_fp16)[name = string("linear_10_cast_fp16")]; string x_29_mode_0 = const()[name = string("x_29_mode_0"), val = string("EXACT")]; tensor x_29_cast_fp16 = gelu(mode = x_29_mode_0, x = linear_10_cast_fp16)[name = string("x_29_cast_fp16")]; tensor var_273_to_fp16 = const()[name = string("op_273_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13869248)))]; tensor var_274_to_fp16 = const()[name = string("op_274_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15966464)))]; tensor linear_11_cast_fp16 = linear(bias = var_274_to_fp16, weight = var_273_to_fp16, x = x_29_cast_fp16)[name = string("linear_11_cast_fp16")]; tensor x_31_cast_fp16 = add(x = x_25_cast_fp16, y = linear_11_cast_fp16)[name = string("x_31_cast_fp16")]; int32 var_284 = const()[name = string("op_284"), val = int32(-1)]; tensor var_300_axes_0 = const()[name = string("op_300_axes_0"), val = tensor([-1])]; tensor blocks_2_attn_ln_weight_to_fp16 = const()[name = string("blocks_2_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15967552)))]; tensor blocks_2_attn_ln_bias_to_fp16 = const()[name = string("blocks_2_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15968640)))]; fp16 var_290_to_fp16 = const()[name = string("op_290_to_fp16"), val = fp16(0x1.5p-17)]; tensor var_300_cast_fp16 = layer_norm(axes = var_300_axes_0, beta = blocks_2_attn_ln_bias_to_fp16, epsilon = var_290_to_fp16, gamma = blocks_2_attn_ln_weight_to_fp16, x = x_31_cast_fp16)[name = string("op_300_cast_fp16")]; tensor var_311_to_fp16 = const()[name = string("op_311_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15969728)))]; tensor var_312_to_fp16 = const()[name = string("op_312_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16494080)))]; tensor linear_12_cast_fp16 = linear(bias = var_312_to_fp16, weight = var_311_to_fp16, x = var_300_cast_fp16)[name = string("linear_12_cast_fp16")]; tensor var_315_to_fp16 = const()[name = string("op_315_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16495168)))]; tensor linear_13_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_315_to_fp16, x = var_300_cast_fp16)[name = string("linear_13_cast_fp16")]; tensor var_319_to_fp16 = const()[name = string("op_319_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17019520)))]; tensor var_320_to_fp16 = const()[name = string("op_320_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17543872)))]; tensor linear_14_cast_fp16 = linear(bias = var_320_to_fp16, weight = var_319_to_fp16, x = var_300_cast_fp16)[name = string("linear_14_cast_fp16")]; tensor var_328 = const()[name = string("op_328"), val = tensor([1, 1500, 8, -1])]; tensor var_329_cast_fp16 = reshape(shape = var_328, x = linear_12_cast_fp16)[name = string("op_329_cast_fp16")]; tensor const_46_to_fp16 = const()[name = string("const_46_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; tensor q_11_cast_fp16 = mul(x = var_329_cast_fp16, y = const_46_to_fp16)[name = string("q_11_cast_fp16")]; tensor var_335 = const()[name = string("op_335"), val = tensor([1, 1500, 8, -1])]; tensor var_336_cast_fp16 = reshape(shape = var_335, x = linear_13_cast_fp16)[name = string("op_336_cast_fp16")]; tensor const_47_to_fp16 = const()[name = string("const_47_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; tensor k_11_cast_fp16 = mul(x = var_336_cast_fp16, y = const_47_to_fp16)[name = string("k_11_cast_fp16")]; tensor var_342 = const()[name = string("op_342"), val = tensor([1, 1500, 8, -1])]; tensor var_343_cast_fp16 = reshape(shape = var_342, x = linear_14_cast_fp16)[name = string("op_343_cast_fp16")]; tensor var_344 = const()[name = string("op_344"), val = tensor([0, 2, 1, 3])]; bool qk_5_transpose_x_0 = const()[name = string("qk_5_transpose_x_0"), val = bool(false)]; bool qk_5_transpose_y_0 = const()[name = string("qk_5_transpose_y_0"), val = bool(false)]; tensor transpose_28_perm_0 = const()[name = string("transpose_28_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_29_perm_0 = const()[name = string("transpose_29_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_29 = transpose(perm = transpose_29_perm_0, x = k_11_cast_fp16)[name = string("transpose_49")]; tensor transpose_28 = transpose(perm = transpose_28_perm_0, x = q_11_cast_fp16)[name = string("transpose_50")]; tensor qk_5_cast_fp16 = matmul(transpose_x = qk_5_transpose_x_0, transpose_y = qk_5_transpose_y_0, x = transpose_28, y = transpose_29)[name = string("qk_5_cast_fp16")]; tensor var_348_cast_fp16 = softmax(axis = var_284, x = qk_5_cast_fp16)[name = string("op_348_cast_fp16")]; bool var_350_transpose_x_0 = const()[name = string("op_350_transpose_x_0"), val = bool(false)]; bool var_350_transpose_y_0 = const()[name = string("op_350_transpose_y_0"), val = bool(false)]; tensor v_11_cast_fp16 = transpose(perm = var_344, x = var_343_cast_fp16)[name = string("transpose_51")]; tensor var_350_cast_fp16 = matmul(transpose_x = var_350_transpose_x_0, transpose_y = var_350_transpose_y_0, x = var_348_cast_fp16, y = v_11_cast_fp16)[name = string("op_350_cast_fp16")]; tensor var_351 = const()[name = string("op_351"), val = tensor([0, 2, 1, 3])]; tensor concat_2 = const()[name = string("concat_2"), val = tensor([1, 1500, 512])]; tensor var_352_cast_fp16 = transpose(perm = var_351, x = var_350_cast_fp16)[name = string("transpose_48")]; tensor x_35_cast_fp16 = reshape(shape = concat_2, x = var_352_cast_fp16)[name = string("x_35_cast_fp16")]; tensor var_356_to_fp16 = const()[name = string("op_356_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17544960)))]; tensor var_357_to_fp16 = const()[name = string("op_357_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18069312)))]; tensor linear_15_cast_fp16 = linear(bias = var_357_to_fp16, weight = var_356_to_fp16, x = x_35_cast_fp16)[name = string("linear_15_cast_fp16")]; tensor x_37_cast_fp16 = add(x = x_31_cast_fp16, y = linear_15_cast_fp16)[name = string("x_37_cast_fp16")]; tensor var_364_axes_0 = const()[name = string("op_364_axes_0"), val = tensor([-1])]; tensor blocks_2_mlp_ln_weight_to_fp16 = const()[name = string("blocks_2_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18070400)))]; tensor blocks_2_mlp_ln_bias_to_fp16 = const()[name = string("blocks_2_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18071488)))]; tensor var_364_cast_fp16 = layer_norm(axes = var_364_axes_0, beta = blocks_2_mlp_ln_bias_to_fp16, epsilon = var_290_to_fp16, gamma = blocks_2_mlp_ln_weight_to_fp16, x = x_37_cast_fp16)[name = string("op_364_cast_fp16")]; tensor var_373_to_fp16 = const()[name = string("op_373_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18072576)))]; tensor var_374_to_fp16 = const()[name = string("op_374_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20169792)))]; tensor linear_16_cast_fp16 = linear(bias = var_374_to_fp16, weight = var_373_to_fp16, x = var_364_cast_fp16)[name = string("linear_16_cast_fp16")]; string x_41_mode_0 = const()[name = string("x_41_mode_0"), val = string("EXACT")]; tensor x_41_cast_fp16 = gelu(mode = x_41_mode_0, x = linear_16_cast_fp16)[name = string("x_41_cast_fp16")]; tensor var_379_to_fp16 = const()[name = string("op_379_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20173952)))]; tensor var_380_to_fp16 = const()[name = string("op_380_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22271168)))]; tensor linear_17_cast_fp16 = linear(bias = var_380_to_fp16, weight = var_379_to_fp16, x = x_41_cast_fp16)[name = string("linear_17_cast_fp16")]; tensor x_43_cast_fp16 = add(x = x_37_cast_fp16, y = linear_17_cast_fp16)[name = string("x_43_cast_fp16")]; int32 var_390 = const()[name = string("op_390"), val = int32(-1)]; tensor var_406_axes_0 = const()[name = string("op_406_axes_0"), val = tensor([-1])]; tensor blocks_3_attn_ln_weight_to_fp16 = const()[name = string("blocks_3_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22272256)))]; tensor blocks_3_attn_ln_bias_to_fp16 = const()[name = string("blocks_3_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22273344)))]; fp16 var_396_to_fp16 = const()[name = string("op_396_to_fp16"), val = fp16(0x1.5p-17)]; tensor var_406_cast_fp16 = layer_norm(axes = var_406_axes_0, beta = blocks_3_attn_ln_bias_to_fp16, epsilon = var_396_to_fp16, gamma = blocks_3_attn_ln_weight_to_fp16, x = x_43_cast_fp16)[name = string("op_406_cast_fp16")]; tensor var_417_to_fp16 = const()[name = string("op_417_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22274432)))]; tensor var_418_to_fp16 = const()[name = string("op_418_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22798784)))]; tensor linear_18_cast_fp16 = linear(bias = var_418_to_fp16, weight = var_417_to_fp16, x = var_406_cast_fp16)[name = string("linear_18_cast_fp16")]; tensor var_421_to_fp16 = const()[name = string("op_421_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22799872)))]; tensor linear_19_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_421_to_fp16, x = var_406_cast_fp16)[name = string("linear_19_cast_fp16")]; tensor var_425_to_fp16 = const()[name = string("op_425_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23324224)))]; tensor var_426_to_fp16 = const()[name = string("op_426_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23848576)))]; tensor linear_20_cast_fp16 = linear(bias = var_426_to_fp16, weight = var_425_to_fp16, x = var_406_cast_fp16)[name = string("linear_20_cast_fp16")]; tensor var_434 = const()[name = string("op_434"), val = tensor([1, 1500, 8, -1])]; tensor var_435_cast_fp16 = reshape(shape = var_434, x = linear_18_cast_fp16)[name = string("op_435_cast_fp16")]; tensor const_48_to_fp16 = const()[name = string("const_48_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; tensor q_15_cast_fp16 = mul(x = var_435_cast_fp16, y = const_48_to_fp16)[name = string("q_15_cast_fp16")]; tensor var_441 = const()[name = string("op_441"), val = tensor([1, 1500, 8, -1])]; tensor var_442_cast_fp16 = reshape(shape = var_441, x = linear_19_cast_fp16)[name = string("op_442_cast_fp16")]; tensor const_49_to_fp16 = const()[name = string("const_49_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; tensor k_15_cast_fp16 = mul(x = var_442_cast_fp16, y = const_49_to_fp16)[name = string("k_15_cast_fp16")]; tensor var_448 = const()[name = string("op_448"), val = tensor([1, 1500, 8, -1])]; tensor var_449_cast_fp16 = reshape(shape = var_448, x = linear_20_cast_fp16)[name = string("op_449_cast_fp16")]; tensor var_450 = const()[name = string("op_450"), val = tensor([0, 2, 1, 3])]; bool qk_7_transpose_x_0 = const()[name = string("qk_7_transpose_x_0"), val = bool(false)]; bool qk_7_transpose_y_0 = const()[name = string("qk_7_transpose_y_0"), val = bool(false)]; tensor transpose_30_perm_0 = const()[name = string("transpose_30_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_31_perm_0 = const()[name = string("transpose_31_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_31 = transpose(perm = transpose_31_perm_0, x = k_15_cast_fp16)[name = string("transpose_45")]; tensor transpose_30 = transpose(perm = transpose_30_perm_0, x = q_15_cast_fp16)[name = string("transpose_46")]; tensor qk_7_cast_fp16 = matmul(transpose_x = qk_7_transpose_x_0, transpose_y = qk_7_transpose_y_0, x = transpose_30, y = transpose_31)[name = string("qk_7_cast_fp16")]; tensor var_454_cast_fp16 = softmax(axis = var_390, x = qk_7_cast_fp16)[name = string("op_454_cast_fp16")]; bool var_456_transpose_x_0 = const()[name = string("op_456_transpose_x_0"), val = bool(false)]; bool var_456_transpose_y_0 = const()[name = string("op_456_transpose_y_0"), val = bool(false)]; tensor v_15_cast_fp16 = transpose(perm = var_450, x = var_449_cast_fp16)[name = string("transpose_47")]; tensor var_456_cast_fp16 = matmul(transpose_x = var_456_transpose_x_0, transpose_y = var_456_transpose_y_0, x = var_454_cast_fp16, y = v_15_cast_fp16)[name = string("op_456_cast_fp16")]; tensor var_457 = const()[name = string("op_457"), val = tensor([0, 2, 1, 3])]; tensor concat_3 = const()[name = string("concat_3"), val = tensor([1, 1500, 512])]; tensor var_458_cast_fp16 = transpose(perm = var_457, x = var_456_cast_fp16)[name = string("transpose_44")]; tensor x_47_cast_fp16 = reshape(shape = concat_3, x = var_458_cast_fp16)[name = string("x_47_cast_fp16")]; tensor var_462_to_fp16 = const()[name = string("op_462_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23849664)))]; tensor var_463_to_fp16 = const()[name = string("op_463_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24374016)))]; tensor linear_21_cast_fp16 = linear(bias = var_463_to_fp16, weight = var_462_to_fp16, x = x_47_cast_fp16)[name = string("linear_21_cast_fp16")]; tensor x_49_cast_fp16 = add(x = x_43_cast_fp16, y = linear_21_cast_fp16)[name = string("x_49_cast_fp16")]; tensor var_470_axes_0 = const()[name = string("op_470_axes_0"), val = tensor([-1])]; tensor blocks_3_mlp_ln_weight_to_fp16 = const()[name = string("blocks_3_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24375104)))]; tensor blocks_3_mlp_ln_bias_to_fp16 = const()[name = string("blocks_3_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24376192)))]; tensor var_470_cast_fp16 = layer_norm(axes = var_470_axes_0, beta = blocks_3_mlp_ln_bias_to_fp16, epsilon = var_396_to_fp16, gamma = blocks_3_mlp_ln_weight_to_fp16, x = x_49_cast_fp16)[name = string("op_470_cast_fp16")]; tensor var_479_to_fp16 = const()[name = string("op_479_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24377280)))]; tensor var_480_to_fp16 = const()[name = string("op_480_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26474496)))]; tensor linear_22_cast_fp16 = linear(bias = var_480_to_fp16, weight = var_479_to_fp16, x = var_470_cast_fp16)[name = string("linear_22_cast_fp16")]; string x_53_mode_0 = const()[name = string("x_53_mode_0"), val = string("EXACT")]; tensor x_53_cast_fp16 = gelu(mode = x_53_mode_0, x = linear_22_cast_fp16)[name = string("x_53_cast_fp16")]; tensor var_485_to_fp16 = const()[name = string("op_485_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26478656)))]; tensor var_486_to_fp16 = const()[name = string("op_486_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28575872)))]; tensor linear_23_cast_fp16 = linear(bias = var_486_to_fp16, weight = var_485_to_fp16, x = x_53_cast_fp16)[name = string("linear_23_cast_fp16")]; tensor x_55_cast_fp16 = add(x = x_49_cast_fp16, y = linear_23_cast_fp16)[name = string("x_55_cast_fp16")]; int32 var_496 = const()[name = string("op_496"), val = int32(-1)]; tensor var_512_axes_0 = const()[name = string("op_512_axes_0"), val = tensor([-1])]; tensor blocks_4_attn_ln_weight_to_fp16 = const()[name = string("blocks_4_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28576960)))]; tensor blocks_4_attn_ln_bias_to_fp16 = const()[name = string("blocks_4_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28578048)))]; fp16 var_502_to_fp16 = const()[name = string("op_502_to_fp16"), val = fp16(0x1.5p-17)]; tensor var_512_cast_fp16 = layer_norm(axes = var_512_axes_0, beta = blocks_4_attn_ln_bias_to_fp16, epsilon = var_502_to_fp16, gamma = blocks_4_attn_ln_weight_to_fp16, x = x_55_cast_fp16)[name = string("op_512_cast_fp16")]; tensor var_523_to_fp16 = const()[name = string("op_523_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28579136)))]; tensor var_524_to_fp16 = const()[name = string("op_524_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29103488)))]; tensor linear_24_cast_fp16 = linear(bias = var_524_to_fp16, weight = var_523_to_fp16, x = var_512_cast_fp16)[name = string("linear_24_cast_fp16")]; tensor var_527_to_fp16 = const()[name = string("op_527_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29104576)))]; tensor linear_25_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_527_to_fp16, x = var_512_cast_fp16)[name = string("linear_25_cast_fp16")]; tensor var_531_to_fp16 = const()[name = string("op_531_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29628928)))]; tensor var_532_to_fp16 = const()[name = string("op_532_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30153280)))]; tensor linear_26_cast_fp16 = linear(bias = var_532_to_fp16, weight = var_531_to_fp16, x = var_512_cast_fp16)[name = string("linear_26_cast_fp16")]; tensor var_540 = const()[name = string("op_540"), val = tensor([1, 1500, 8, -1])]; tensor var_541_cast_fp16 = reshape(shape = var_540, x = linear_24_cast_fp16)[name = string("op_541_cast_fp16")]; tensor const_50_to_fp16 = const()[name = string("const_50_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; tensor q_19_cast_fp16 = mul(x = var_541_cast_fp16, y = const_50_to_fp16)[name = string("q_19_cast_fp16")]; tensor var_547 = const()[name = string("op_547"), val = tensor([1, 1500, 8, -1])]; tensor var_548_cast_fp16 = reshape(shape = var_547, x = linear_25_cast_fp16)[name = string("op_548_cast_fp16")]; tensor const_51_to_fp16 = const()[name = string("const_51_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; tensor k_19_cast_fp16 = mul(x = var_548_cast_fp16, y = const_51_to_fp16)[name = string("k_19_cast_fp16")]; tensor var_554 = const()[name = string("op_554"), val = tensor([1, 1500, 8, -1])]; tensor var_555_cast_fp16 = reshape(shape = var_554, x = linear_26_cast_fp16)[name = string("op_555_cast_fp16")]; tensor var_556 = const()[name = string("op_556"), val = tensor([0, 2, 1, 3])]; bool qk_9_transpose_x_0 = const()[name = string("qk_9_transpose_x_0"), val = bool(false)]; bool qk_9_transpose_y_0 = const()[name = string("qk_9_transpose_y_0"), val = bool(false)]; tensor transpose_32_perm_0 = const()[name = string("transpose_32_perm_0"), val = tensor([0, 2, -3, -1])]; tensor transpose_33_perm_0 = const()[name = string("transpose_33_perm_0"), val = tensor([0, 2, -1, -3])]; tensor transpose_33 = transpose(perm = transpose_33_perm_0, x = k_19_cast_fp16)[name = string("transpose_41")]; tensor transpose_32 = transpose(perm = transpose_32_perm_0, x = q_19_cast_fp16)[name = string("transpose_42")]; tensor qk_9_cast_fp16 = matmul(transpose_x = qk_9_transpose_x_0, transpose_y = qk_9_transpose_y_0, x = transpose_32, y = transpose_33)[name = string("qk_9_cast_fp16")]; tensor var_560_cast_fp16 = softmax(axis = var_496, x = qk_9_cast_fp16)[name = string("op_560_cast_fp16")]; bool var_562_transpose_x_0 = const()[name = string("op_562_transpose_x_0"), val = bool(false)]; bool var_562_transpose_y_0 = const()[name = string("op_562_transpose_y_0"), val = bool(false)]; tensor v_19_cast_fp16 = transpose(perm = var_556, x = var_555_cast_fp16)[name = string("transpose_43")]; tensor var_562_cast_fp16 = matmul(transpose_x = var_562_transpose_x_0, transpose_y = var_562_transpose_y_0, x = var_560_cast_fp16, y = v_19_cast_fp16)[name = string("op_562_cast_fp16")]; tensor var_563 = const()[name = string("op_563"), val = tensor([0, 2, 1, 3])]; tensor concat_4 = const()[name = string("concat_4"), val = tensor([1, 1500, 512])]; tensor var_564_cast_fp16 = transpose(perm = var_563, x = var_562_cast_fp16)[name = string("transpose_40")]; tensor x_59_cast_fp16 = reshape(shape = concat_4, x = var_564_cast_fp16)[name = string("x_59_cast_fp16")]; tensor