moe+asq (#1)

- upload asq moe quantized model (c01fcbe0d7e2c0d3b6db2d4d326600190b2fab63)

config.json

@@ -0,0 +1,454 @@
+{
+  "architectures": [
+    "DeepseekV3ForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "auto_map": {
+    "AutoConfig": "configuration_deepseek.DeepseekV3Config",
+    "AutoModel": "modeling_deepseek.DeepseekV3Model",
+    "AutoModelForCausalLM": "modeling_deepseek.DeepseekV3ForCausalLM"
+  },
+  "aux_loss_alpha": 0.001,
+  "bos_token_id": 0,
+  "eos_token_id": 1,
+  "ep_size": 1,
+  "first_k_dense_replace": 3,
+  "hidden_act": "silu",
+  "hidden_size": 7168,
+  "initializer_range": 0.02,
+  "intermediate_size": 18432,
+  "kv_lora_rank": 512,
+  "max_position_embeddings": 163840,
+  "model_type": "deepseek_v3",
+  "moe_intermediate_size": 2048,
+  "moe_layer_freq": 1,
+  "n_group": 8,
+  "n_routed_experts": 256,
+  "n_shared_experts": 1,
+  "norm_topk_prob": true,
+  "num_attention_heads": 128,
+  "num_experts_per_tok": 8,
+  "num_hidden_layers": 61,
+  "num_key_value_heads": 128,
+  "num_nextn_predict_layers": 1,
+  "pad_token_id": 128815,
+  "pretraining_tp": 1,
+  "q_lora_rank": 1536,
+  "qk_nope_head_dim": 128,
+  "qk_rope_head_dim": 64,
+  "quantization_config": {
+    "algo_config": {
+      "compute_scale_loss": "MAE",
+      "model_decoder_layers": "model.layers",
+      "name": "autosmoothquant",
+      "num_attention_heads": -1,
+      "num_key_value_heads": -1,
+      "scaling_layers": [
+        {
+          "inp": "mlp.gate_proj",
+          "layers": [
+            "mlp.gate_proj",
+            "mlp.up_proj"
+          ],
+          "module2inspect": "mlp",
+          "prev_op": "post_attention_layernorm"
+        },
+        {
+          "inp": "mlp.down_proj",
+          "layers": [
+            "mlp.down_proj"
+          ],
+          "prev_op": "mlp.up_proj"
+        },
+        {
+          "inp": "down_proj",
+          "layers": [
+            "down_proj"
+          ],
+          "prev_op": "up_proj"
+        }
+      ]
+    },
+    "exclude": [
+      "model.layers.0.self_attn.q_a_proj",
+      "model.layers.0.self_attn.q_b_proj",
+      "model.layers.0.self_attn.kv_a_proj_with_mqa",
+      "model.layers.0.self_attn.kv_b_proj",
+      "model.layers.0.self_attn.o_proj",
+      "model.layers.1.self_attn.q_a_proj",
+      "model.layers.1.self_attn.q_b_proj",
+      "model.layers.1.self_attn.kv_a_proj_with_mqa",
+      "model.layers.1.self_attn.kv_b_proj",
+      "model.layers.1.self_attn.o_proj",
+      "model.layers.2.self_attn.q_a_proj",
+      "model.layers.2.self_attn.q_b_proj",
+      "model.layers.2.self_attn.kv_a_proj_with_mqa",
+      "model.layers.2.self_attn.kv_b_proj",
+      "model.layers.2.self_attn.o_proj",
+      "model.layers.3.self_attn.q_a_proj",
+      "model.layers.3.self_attn.q_b_proj",
+      "model.layers.3.self_attn.kv_a_proj_with_mqa",
+      "model.layers.3.self_attn.kv_b_proj",
+      "model.layers.3.self_attn.o_proj",
+      "model.layers.4.self_attn.q_a_proj",
+      "model.layers.4.self_attn.q_b_proj",
+      "model.layers.4.self_attn.kv_a_proj_with_mqa",
+      "model.layers.4.self_attn.kv_b_proj",
+      "model.layers.4.self_attn.o_proj",
+      "model.layers.5.self_attn.q_a_proj",
+      "model.layers.5.self_attn.q_b_proj",
+      "model.layers.5.self_attn.kv_a_proj_with_mqa",
+      "model.layers.5.self_attn.kv_b_proj",
+      "model.layers.5.self_attn.o_proj",
+      "model.layers.6.self_attn.q_a_proj",
+      "model.layers.6.self_attn.q_b_proj",
+      "model.layers.6.self_attn.kv_a_proj_with_mqa",
+      "model.layers.6.self_attn.kv_b_proj",
+      "model.layers.6.self_attn.o_proj",
+      "model.layers.7.self_attn.q_a_proj",
+      "model.layers.7.self_attn.q_b_proj",
+      "model.layers.7.self_attn.kv_a_proj_with_mqa",
+      "model.layers.7.self_attn.kv_b_proj",
+      "model.layers.7.self_attn.o_proj",
+      "model.layers.8.self_attn.q_a_proj",
+      "model.layers.8.self_attn.q_b_proj",
+      "model.layers.8.self_attn.kv_a_proj_with_mqa",
+      "model.layers.8.self_attn.kv_b_proj",
+      "model.layers.8.self_attn.o_proj",
+      "model.layers.9.self_attn.q_a_proj",
+      "model.layers.9.self_attn.q_b_proj",
+      "model.layers.9.self_attn.kv_a_proj_with_mqa",
+      "model.layers.9.self_attn.kv_b_proj",
+      "model.layers.9.self_attn.o_proj",
+      "model.layers.10.self_attn.q_a_proj",
+      "model.layers.10.self_attn.q_b_proj",
+      "model.layers.10.self_attn.kv_a_proj_with_mqa",
+      "model.layers.10.self_attn.kv_b_proj",
+      "model.layers.10.self_attn.o_proj",
+      "model.layers.11.self_attn.q_a_proj",
+      "model.layers.11.self_attn.q_b_proj",
+      "model.layers.11.self_attn.kv_a_proj_with_mqa",
+      "model.layers.11.self_attn.kv_b_proj",
+      "model.layers.11.self_attn.o_proj",
+      "model.layers.12.self_attn.q_a_proj",
+      "model.layers.12.self_attn.q_b_proj",
+      "model.layers.12.self_attn.kv_a_proj_with_mqa",
+      "model.layers.12.self_attn.kv_b_proj",
+      "model.layers.12.self_attn.o_proj",
+      "model.layers.13.self_attn.q_a_proj",
+      "model.layers.13.self_attn.q_b_proj",
+      "model.layers.13.self_attn.kv_a_proj_with_mqa",
+      "model.layers.13.self_attn.kv_b_proj",
+      "model.layers.13.self_attn.o_proj",
+      "model.layers.14.self_attn.q_a_proj",
+      "model.layers.14.self_attn.q_b_proj",
+      "model.layers.14.self_attn.kv_a_proj_with_mqa",
+      "model.layers.14.self_attn.kv_b_proj",
+      "model.layers.14.self_attn.o_proj",
+      "model.layers.15.self_attn.q_a_proj",
+      "model.layers.15.self_attn.q_b_proj",
+      "model.layers.15.self_attn.kv_a_proj_with_mqa",
+      "model.layers.15.self_attn.kv_b_proj",
+      "model.layers.15.self_attn.o_proj",
+      "model.layers.16.self_attn.q_a_proj",
+      "model.layers.16.self_attn.q_b_proj",
+      "model.layers.16.self_attn.kv_a_proj_with_mqa",
+      "model.layers.16.self_attn.kv_b_proj",
+      "model.layers.16.self_attn.o_proj",
+      "model.layers.17.self_attn.q_a_proj",
+      "model.layers.17.self_attn.q_b_proj",
+      "model.layers.17.self_attn.kv_a_proj_with_mqa",
+      "model.layers.17.self_attn.kv_b_proj",
+      "model.layers.17.self_attn.o_proj",
+      "model.layers.18.self_attn.q_a_proj",
+      "model.layers.18.self_attn.q_b_proj",
+      "model.layers.18.self_attn.kv_a_proj_with_mqa",
+      "model.layers.18.self_attn.kv_b_proj",
+      "model.layers.18.self_attn.o_proj",
+      "model.layers.19.self_attn.q_a_proj",
+      "model.layers.19.self_attn.q_b_proj",
+      "model.layers.19.self_attn.kv_a_proj_with_mqa",
+      "model.layers.19.self_attn.kv_b_proj",
+      "model.layers.19.self_attn.o_proj",
+      "model.layers.20.self_attn.q_a_proj",
+      "model.layers.20.self_attn.q_b_proj",
+      "model.layers.20.self_attn.kv_a_proj_with_mqa",
+      "model.layers.20.self_attn.kv_b_proj",
+      "model.layers.20.self_attn.o_proj",
+      "model.layers.21.self_attn.q_a_proj",
+      "model.layers.21.self_attn.q_b_proj",
+      "model.layers.21.self_attn.kv_a_proj_with_mqa",
+      "model.layers.21.self_attn.kv_b_proj",
+      "model.layers.21.self_attn.o_proj",
+      "model.layers.22.self_attn.q_a_proj",
+      "model.layers.22.self_attn.q_b_proj",
+      "model.layers.22.self_attn.kv_a_proj_with_mqa",
+      "model.layers.22.self_attn.kv_b_proj",
+      "model.layers.22.self_attn.o_proj",
+      "model.layers.23.self_attn.q_a_proj",
+      "model.layers.23.self_attn.q_b_proj",
+      "model.layers.23.self_attn.kv_a_proj_with_mqa",
+      "model.layers.23.self_attn.kv_b_proj",
+      "model.layers.23.self_attn.o_proj",
+      "model.layers.24.self_attn.q_a_proj",
+      "model.layers.24.self_attn.q_b_proj",
+      "model.layers.24.self_attn.kv_a_proj_with_mqa",
+      "model.layers.24.self_attn.kv_b_proj",
+      "model.layers.24.self_attn.o_proj",
+      "model.layers.25.self_attn.q_a_proj",
+      "model.layers.25.self_attn.q_b_proj",
+      "model.layers.25.self_attn.kv_a_proj_with_mqa",
+      "model.layers.25.self_attn.kv_b_proj",
+      "model.layers.25.self_attn.o_proj",
+      "model.layers.26.self_attn.q_a_proj",
+      "model.layers.26.self_attn.q_b_proj",
+      "model.layers.26.self_attn.kv_a_proj_with_mqa",
+      "model.layers.26.self_attn.kv_b_proj",
+      "model.layers.26.self_attn.o_proj",
+      "model.layers.27.self_attn.q_a_proj",
+      "model.layers.27.self_attn.q_b_proj",
+      "model.layers.27.self_attn.kv_a_proj_with_mqa",
+      "model.layers.27.self_attn.kv_b_proj",
+      "model.layers.27.self_attn.o_proj",
+      "model.layers.28.self_attn.q_a_proj",
+      "model.layers.28.self_attn.q_b_proj",
+      "model.layers.28.self_attn.kv_a_proj_with_mqa",
+      "model.layers.28.self_attn.kv_b_proj",
+      "model.layers.28.self_attn.o_proj",
+      "model.layers.29.self_attn.q_a_proj",
+      "model.layers.29.self_attn.q_b_proj",
+      "model.layers.29.self_attn.kv_a_proj_with_mqa",
+      "model.layers.29.self_attn.kv_b_proj",
+      "model.layers.29.self_attn.o_proj",
+      "model.layers.30.self_attn.q_a_proj",
+      "model.layers.30.self_attn.q_b_proj",
+      "model.layers.30.self_attn.kv_a_proj_with_mqa",
+      "model.layers.30.self_attn.kv_b_proj",
+      "model.layers.30.self_attn.o_proj",
+      "model.layers.31.self_attn.q_a_proj",
+      "model.layers.31.self_attn.q_b_proj",
+      "model.layers.31.self_attn.kv_a_proj_with_mqa",
+      "model.layers.31.self_attn.kv_b_proj",
+      "model.layers.31.self_attn.o_proj",
+      "model.layers.32.self_attn.q_a_proj",
+      "model.layers.32.self_attn.q_b_proj",
+      "model.layers.32.self_attn.kv_a_proj_with_mqa",
+      "model.layers.32.self_attn.kv_b_proj",
+      "model.layers.32.self_attn.o_proj",
+      "model.layers.33.self_attn.q_a_proj",
+      "model.layers.33.self_attn.q_b_proj",
+      "model.layers.33.self_attn.kv_a_proj_with_mqa",
+      "model.layers.33.self_attn.kv_b_proj",
+      "model.layers.33.self_attn.o_proj",
+      "model.layers.34.self_attn.q_a_proj",
+      "model.layers.34.self_attn.q_b_proj",
+      "model.layers.34.self_attn.kv_a_proj_with_mqa",
+      "model.layers.34.self_attn.kv_b_proj",
+      "model.layers.34.self_attn.o_proj",
+      "model.layers.35.self_attn.q_a_proj",
+      "model.layers.35.self_attn.q_b_proj",
+      "model.layers.35.self_attn.kv_a_proj_with_mqa",
+      "model.layers.35.self_attn.kv_b_proj",
+      "model.layers.35.self_attn.o_proj",
+      "model.layers.36.self_attn.q_a_proj",
+      "model.layers.36.self_attn.q_b_proj",
+      "model.layers.36.self_attn.kv_a_proj_with_mqa",
+      "model.layers.36.self_attn.kv_b_proj",
+      "model.layers.36.self_attn.o_proj",
+      "model.layers.37.self_attn.q_a_proj",
+      "model.layers.37.self_attn.q_b_proj",
+      "model.layers.37.self_attn.kv_a_proj_with_mqa",
+      "model.layers.37.self_attn.kv_b_proj",
+      "model.layers.37.self_attn.o_proj",
+      "model.layers.38.self_attn.q_a_proj",
+      "model.layers.38.self_attn.q_b_proj",
+      "model.layers.38.self_attn.kv_a_proj_with_mqa",
+      "model.layers.38.self_attn.kv_b_proj",
+      "model.layers.38.self_attn.o_proj",
+      "model.layers.39.self_attn.q_a_proj",
+      "model.layers.39.self_attn.q_b_proj",
+      "model.layers.39.self_attn.kv_a_proj_with_mqa",
+      "model.layers.39.self_attn.kv_b_proj",
+      "model.layers.39.self_attn.o_proj",
+      "model.layers.40.self_attn.q_a_proj",
+      "model.layers.40.self_attn.q_b_proj",
+      "model.layers.40.self_attn.kv_a_proj_with_mqa",
+      "model.layers.40.self_attn.kv_b_proj",
+      "model.layers.40.self_attn.o_proj",
+      "model.layers.41.self_attn.q_a_proj",
+      "model.layers.41.self_attn.q_b_proj",
+      "model.layers.41.self_attn.kv_a_proj_with_mqa",
+      "model.layers.41.self_attn.kv_b_proj",
+      "model.layers.41.self_attn.o_proj",
+      "model.layers.42.self_attn.q_a_proj",
+      "model.layers.42.self_attn.q_b_proj",
+      "model.layers.42.self_attn.kv_a_proj_with_mqa",
+      "model.layers.42.self_attn.kv_b_proj",
+      "model.layers.42.self_attn.o_proj",
+      "model.layers.43.self_attn.q_a_proj",
+      "model.layers.43.self_attn.q_b_proj",
+      "model.layers.43.self_attn.kv_a_proj_with_mqa",
+      "model.layers.43.self_attn.kv_b_proj",
+      "model.layers.43.self_attn.o_proj",
+      "model.layers.44.self_attn.q_a_proj",
+      "model.layers.44.self_attn.q_b_proj",
+      "model.layers.44.self_attn.kv_a_proj_with_mqa",
+      "model.layers.44.self_attn.kv_b_proj",
+      "model.layers.44.self_attn.o_proj",
+      "model.layers.45.self_attn.q_a_proj",
+      "model.layers.45.self_attn.q_b_proj",
+      "model.layers.45.self_attn.kv_a_proj_with_mqa",
+      "model.layers.45.self_attn.kv_b_proj",
+      "model.layers.45.self_attn.o_proj",
+      "model.layers.46.self_attn.q_a_proj",
+      "model.layers.46.self_attn.q_b_proj",
+      "model.layers.46.self_attn.kv_a_proj_with_mqa",
+      "model.layers.46.self_attn.kv_b_proj",
+      "model.layers.46.self_attn.o_proj",
+      "model.layers.47.self_attn.q_a_proj",
+      "model.layers.47.self_attn.q_b_proj",
+      "model.layers.47.self_attn.kv_a_proj_with_mqa",
+      "model.layers.47.self_attn.kv_b_proj",
+      "model.layers.47.self_attn.o_proj",
+      "model.layers.48.self_attn.q_a_proj",
+      "model.layers.48.self_attn.q_b_proj",
+      "model.layers.48.self_attn.kv_a_proj_with_mqa",
+      "model.layers.48.self_attn.kv_b_proj",
+      "model.layers.48.self_attn.o_proj",
+      "model.layers.49.self_attn.q_a_proj",
+      "model.layers.49.self_attn.q_b_proj",
+      "model.layers.49.self_attn.kv_a_proj_with_mqa",
+      "model.layers.49.self_attn.kv_b_proj",
+      "model.layers.49.self_attn.o_proj",
+      "model.layers.50.self_attn.q_a_proj",
+      "model.layers.50.self_attn.q_b_proj",
+      "model.layers.50.self_attn.kv_a_proj_with_mqa",
+      "model.layers.50.self_attn.kv_b_proj",
+      "model.layers.50.self_attn.o_proj",
+      "model.layers.51.self_attn.q_a_proj",
+      "model.layers.51.self_attn.q_b_proj",
+      "model.layers.51.self_attn.kv_a_proj_with_mqa",
+      "model.layers.51.self_attn.kv_b_proj",
+      "model.layers.51.self_attn.o_proj",
+      "model.layers.52.self_attn.q_a_proj",
+      "model.layers.52.self_attn.q_b_proj",
+      "model.layers.52.self_attn.kv_a_proj_with_mqa",
+      "model.layers.52.self_attn.kv_b_proj",
+      "model.layers.52.self_attn.o_proj",
+      "model.layers.53.self_attn.q_a_proj",
+      "model.layers.53.self_attn.q_b_proj",
+      "model.layers.53.self_attn.kv_a_proj_with_mqa",
+      "model.layers.53.self_attn.kv_b_proj",
+      "model.layers.53.self_attn.o_proj",
+      "model.layers.54.self_attn.q_a_proj",
+      "model.layers.54.self_attn.q_b_proj",
+      "model.layers.54.self_attn.kv_a_proj_with_mqa",
+      "model.layers.54.self_attn.kv_b_proj",
+      "model.layers.54.self_attn.o_proj",
+      "model.layers.55.self_attn.q_a_proj",
+      "model.layers.55.self_attn.q_b_proj",
+      "model.layers.55.self_attn.kv_a_proj_with_mqa",
+      "model.layers.55.self_attn.kv_b_proj",
+      "model.layers.55.self_attn.o_proj",
+      "model.layers.56.self_attn.q_a_proj",
+      "model.layers.56.self_attn.q_b_proj",
+      "model.layers.56.self_attn.kv_a_proj_with_mqa",
+      "model.layers.56.self_attn.kv_b_proj",
+      "model.layers.56.self_attn.o_proj",
+      "model.layers.57.self_attn.q_a_proj",
+      "model.layers.57.self_attn.q_b_proj",
+      "model.layers.57.self_attn.kv_a_proj_with_mqa",
+      "model.layers.57.self_attn.kv_b_proj",
+      "model.layers.57.self_attn.o_proj",
+      "model.layers.58.self_attn.q_a_proj",
+      "model.layers.58.self_attn.q_b_proj",
+      "model.layers.58.self_attn.kv_a_proj_with_mqa",
+      "model.layers.58.self_attn.kv_b_proj",
+      "model.layers.58.self_attn.o_proj",
+      "model.layers.59.self_attn.q_a_proj",
+      "model.layers.59.self_attn.q_b_proj",
+      "model.layers.59.self_attn.kv_a_proj_with_mqa",
+      "model.layers.59.self_attn.kv_b_proj",
+      "model.layers.59.self_attn.o_proj",
+      "model.layers.60.self_attn.q_a_proj",
+      "model.layers.60.self_attn.q_b_proj",
+      "model.layers.60.self_attn.kv_a_proj_with_mqa",
+      "model.layers.60.self_attn.kv_b_proj",
+      "model.layers.60.self_attn.o_proj",
+      "lm_head"
+    ],
+    "export": {
+      "kv_cache_group": [],
+      "min_kv_scale": 0.0,
+      "pack_method": "reorder",
+      "weight_format": "real_quantized",
+      "weight_merge_groups": null
+    },
+    "global_quant_config": {
+      "bias": null,
+      "input_tensors": {
+        "ch_axis": -1,
+        "dtype": "fp4",
+        "group_size": 32,
+        "is_dynamic": true,
+        "is_mx_scale_constraint": false,
+        "is_scale_quant": false,
+        "mx_element_dtype": null,
+        "observer_cls": "PerBlockMXObserver",
+        "qscheme": "per_group",
+        "round_method": "half_even",
+        "scale_calculation_mode": "even",
+        "scale_format": "e8m0",
+        "scale_type": "float",
+        "symmetric": null
+      },
+      "output_tensors": null,
+      "target_device": null,
+      "weight": {
+        "ch_axis": -1,
+        "dtype": "fp4",
+        "group_size": 32,
+        "is_dynamic": false,
+        "is_mx_scale_constraint": false,
+        "is_scale_quant": false,
+        "mx_element_dtype": null,
+        "observer_cls": "PerBlockMXObserver",
+        "qscheme": "per_group",
+        "round_method": "half_even",
+        "scale_calculation_mode": "even",
+        "scale_format": "e8m0",
+        "scale_type": "float",
+        "symmetric": null
+      }
+    },
+    "layer_quant_config": {},
+    "layer_type_quant_config": {},
+    "quant_method": "quark",
+    "quant_mode": "eager_mode",
+    "softmax_quant_spec": null
+  },
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": {
+    "beta_fast": 32,
+    "beta_slow": 1,
+    "factor": 40,
+    "mscale": 1.0,
+    "mscale_all_dim": 1.0,
+    "original_max_position_embeddings": 4096,
+    "type": "yarn"
+  },
+  "rope_theta": 10000,
+  "routed_scaling_factor": 2.5,
+  "scoring_func": "sigmoid",
+  "seq_aux": true,
+  "tie_word_embeddings": false,
+  "topk_group": 4,
+  "topk_method": "noaux_tc",
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.51.3",
+  "unsloth_fixed": true,
+  "use_cache": true,
+  "v_head_dim": 128,
+  "vocab_size": 129280
+}

configuration_deepseek.py

@@ -0,0 +1,210 @@
+from transformers.configuration_utils import PretrainedConfig
+from transformers.utils import logging
+
+logger = logging.get_logger(__name__)
+
+DEEPSEEK_PRETRAINED_CONFIG_ARCHIVE_MAP = {}
+class DeepseekV3Config(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`DeepseekV3Model`]. It is used to instantiate an DeepSeek
+    model according to the specified arguments, defining the model architecture. Instantiating a configuration with the
+    defaults will yield a similar configuration to that of the DeepSeek-V3.
+
+    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
+    documentation from [`PretrainedConfig`] for more information.
+
+
+    Args:
+        vocab_size (`int`, *optional*, defaults to 129280):
+            Vocabulary size of the Deep model. Defines the number of different tokens that can be represented by the
+            `inputs_ids` passed when calling [`DeepseekV3Model`]
+        hidden_size (`int`, *optional*, defaults to 4096):
+            Dimension of the hidden representations.
+        intermediate_size (`int`, *optional*, defaults to 11008):
+            Dimension of the MLP representations.
+        moe_intermediate_size (`int`, *optional*, defaults to 1407):
+            Dimension of the MoE representations.
+        num_hidden_layers (`int`, *optional*, defaults to 32):
+            Number of hidden layers in the Transformer decoder.
+        num_nextn_predict_layers (`int`, *optional*, defaults to 1):
+            Number of nextn predict layers in the DeepSeekV3 Model.
+        num_attention_heads (`int`, *optional*, defaults to 32):
+            Number of attention heads for each attention layer in the Transformer decoder.
+        n_shared_experts (`int`, *optional*, defaults to None):
+            Number of shared experts, None means dense model.
+        n_routed_experts (`int`, *optional*, defaults to None):
+            Number of routed experts, None means dense model.
+        routed_scaling_factor (`float`, *optional*, defaults to 1.0):
+            Scaling factor or routed experts.
+        topk_method (`str`, *optional*, defaults to `gready`):
+            Topk method used in routed gate.
+        n_group (`int`, *optional*, defaults to None):
+            Number of groups for routed experts.
+        topk_group (`int`, *optional*, defaults to None):
+            Number of selected groups for each token(for each token, ensuring the selected experts is only within `topk_group` groups).
+        num_experts_per_tok (`int`, *optional*, defaults to None):
+            Number of selected experts, None means dense model.
+        moe_layer_freq (`int`, *optional*, defaults to 1):
+            The frequency of the MoE layer: one expert layer for every `moe_layer_freq - 1` dense layers.
+        first_k_dense_replace (`int`, *optional*, defaults to 0):
+            Number of dense layers in shallow layers(embed->dense->dense->...->dense->moe->moe...->lm_head).
+                                                            \--k dense layers--/
+        norm_topk_prob (`bool`, *optional*, defaults to False):
+            Whether to normalize the weights of the routed experts.
+        scoring_func (`str`, *optional*, defaults to 'softmax'):
+            Method of computing expert weights.
+        aux_loss_alpha (`float`, *optional*, defaults to 0.001):
+            Auxiliary loss weight coefficient.
+        seq_aux = (`bool`, *optional*, defaults to True):
+            Whether to compute the auxiliary loss for each individual sample.
+        num_key_value_heads (`int`, *optional*):
+            This is the number of key_value heads that should be used to implement Grouped Query Attention. If
+            `num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if
+            `num_key_value_heads=1 the model will use Multi Query Attention (MQA) otherwise GQA is used. When
+            converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed
+            by meanpooling all the original heads within that group. For more details checkout [this
+            paper](https://arxiv.org/pdf/2305.13245.pdf). If it is not specified, will default to
+            `num_attention_heads`.
+        hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
+            The non-linear activation function (function or string) in the decoder.
+        max_position_embeddings (`int`, *optional*, defaults to 2048):
+            The maximum sequence length that this model might ever be used with.
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        rms_norm_eps (`float`, *optional*, defaults to 1e-06):
+            The epsilon used by the rms normalization layers.
+        use_cache (`bool`, *optional*, defaults to `True`):
+            Whether or not the model should return the last key/values attentions (not used by all models). Only
+            relevant if `config.is_decoder=True`.
+        pad_token_id (`int`, *optional*):
+            Padding token id.
+        bos_token_id (`int`, *optional*, defaults to 1):
+            Beginning of stream token id.
+        eos_token_id (`int`, *optional*, defaults to 2):
+            End of stream token id.
+        pretraining_tp (`int`, *optional*, defaults to 1):
+            Experimental feature. Tensor parallelism rank used during pretraining. Please refer to [this
+            document](https://huggingface.co/docs/transformers/parallelism) to understand more about it. This value is
+            necessary to ensure exact reproducibility of the pretraining results. Please refer to [this
+            issue](https://github.com/pytorch/pytorch/issues/76232).
+        tie_word_embeddings (`bool`, *optional*, defaults to `False`):
+            Whether to tie weight embeddings
+        rope_theta (`float`, *optional*, defaults to 10000.0):
+            The base period of the RoPE embeddings.
+        rope_scaling (`Dict`, *optional*):
+            Dictionary containing the scaling configuration for the RoPE embeddings. Currently supports two scaling
+            strategies: linear and dynamic. Their scaling factor must be a float greater than 1. The expected format is
+            `{"type": strategy name, "factor": scaling factor}`. When using this flag, don't update
+            `max_position_embeddings` to the expected new maximum.
+        attention_bias (`bool`, defaults to `False`, *optional*, defaults to `False`):
+            Whether to use a bias in the query, key, value and output projection layers during self-attention.
+        attention_dropout (`float`, *optional*, defaults to 0.0):
+            The dropout ratio for the attention probabilities.
+
+    ```python
+    >>> from transformers import DeepseekV3Model, DeepseekV3Config
+
+    >>> # Initializing a Deepseek-V3 style configuration
+    >>> configuration = DeepseekV3Config()
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+
+    model_type = "deepseek_v3"
+    keys_to_ignore_at_inference = ["past_key_values"]
+
+    def __init__(
+        self,
+        vocab_size=129280,
+        hidden_size=7168,
+        intermediate_size=18432,
+        moe_intermediate_size = 2048,
+        num_hidden_layers=61,
+        num_nextn_predict_layers=1,
+        num_attention_heads=128,
+        num_key_value_heads=128,
+        n_shared_experts = 1,
+        n_routed_experts = 256,
+        ep_size = 1,
+        routed_scaling_factor = 2.5,
+        kv_lora_rank = 512,
+        q_lora_rank = 1536,
+        qk_rope_head_dim = 64,
+        v_head_dim = 128,
+        qk_nope_head_dim = 128,
+        topk_method = 'noaux_tc',
+        n_group = 8,
+        topk_group = 4,
+        num_experts_per_tok = 8,
+        moe_layer_freq = 1,
+        first_k_dense_replace = 3,
+        norm_topk_prob = True,
+        scoring_func = 'sigmoid',
+        aux_loss_alpha = 0.001,
+        seq_aux = True,
+        hidden_act="silu",
+        max_position_embeddings=4096,
+        initializer_range=0.02,
+        rms_norm_eps=1e-6,
+        use_cache=True,
+        pad_token_id=None,
+        bos_token_id=0,
+        eos_token_id=1,
+        pretraining_tp=1,
+        tie_word_embeddings=False,
+        rope_theta=10000.0,
+        rope_scaling=None,
+        attention_bias=False,
+        attention_dropout=0.0,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.max_position_embeddings = max_position_embeddings
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.moe_intermediate_size = moe_intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_nextn_predict_layers = num_nextn_predict_layers
+        self.num_attention_heads = num_attention_heads
+        self.n_shared_experts = n_shared_experts
+        self.n_routed_experts = n_routed_experts
+        self.ep_size = ep_size
+        self.routed_scaling_factor = routed_scaling_factor
+        self.kv_lora_rank = kv_lora_rank
+        self.q_lora_rank = q_lora_rank
+        self.qk_rope_head_dim = qk_rope_head_dim
+        self.v_head_dim = v_head_dim
+        self.qk_nope_head_dim = qk_nope_head_dim
+        self.topk_method = topk_method
+        self.n_group = n_group
+        self.topk_group = topk_group
+        self.num_experts_per_tok = num_experts_per_tok
+        self.moe_layer_freq = moe_layer_freq
+        self.first_k_dense_replace = first_k_dense_replace
+        self.norm_topk_prob = norm_topk_prob
+        self.scoring_func = scoring_func
+        self.aux_loss_alpha = aux_loss_alpha
+        self.seq_aux = seq_aux
+        # for backward compatibility
+        if num_key_value_heads is None:
+            num_key_value_heads = num_attention_heads
+
+        self.num_key_value_heads = num_key_value_heads
+        self.hidden_act = hidden_act
+        self.initializer_range = initializer_range
+        self.rms_norm_eps = rms_norm_eps
+        self.pretraining_tp = pretraining_tp
+        self.use_cache = use_cache
+        self.rope_theta = rope_theta
+        self.rope_scaling = rope_scaling
+        self.attention_bias = attention_bias
+        self.attention_dropout = attention_dropout
+
+        super().__init__(
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )

generation_config.json

@@ -0,0 +1,9 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 0,
+  "do_sample": true,
+  "eos_token_id": 1,
+  "temperature": 0.6,
+  "top_p": 0.95,
+  "transformers_version": "4.51.3"
+}

model-00001-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:63ca4d560e0faef0df165cbe23501a40af48b3e6e0af1ef71e9c0e681ff46d40
+size 4995884768

model-00002-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f4ca2ac91eadcb0941d6feb5d25100c5dccd0ad78a872e0173900adff2de0d36
+size 4995052496

model-00003-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:48fe7856e8527ba5efd6c086aa0d15aefa7e0b4ac37785cedf72f6512db1da57
+size 4999071368

model-00004-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e201d7a764d220b10fabaf9f3af1a69b365222a7b2b426aa58ef0a4ec69815cf
+size 4994943096

model-00005-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d95eecd8de3bd59a037d882568bd947b10159fc4ef23dec19cb0625af0a60ad5
+size 4994943232

model-00006-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bc68ae966c0d2f2cd737e24dddb2bf17bac5e8727295ca25f25989bd55123d2a
+size 4994943456

model-00007-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:958675a01ba08cf5231a886ffa1b40d12105b13df6ebedebf58bcd62fb015963
+size 4999180624

model-00008-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1e679ba36e007c55aa86f671f3e5758357c17a84a5b04600eeb035daa12be832
+size 4994943096

model-00009-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6c5e2ba4d9c18572331e31e3f09f24aeebf806434db2a8ce890643248cf3a91f
+size 4994943096

model-00010-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0fecc475b57c858f8be5a82e7055572f2fe1db646bb3b8bee892c0822e8f2cca
+size 4994943720

model-00011-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2a9df0953d2a655283ac4321a52cff7ba8e6e7bfcce3fd4e49e071154baef4ca
+size 4916044152

model-00012-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a0b324256e9900992b5718dbc016fac1c9df249cb68f5a46aec47d6d47ba805f
+size 5000091936

model-00013-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:da266c07535de2820652b58a52a2f556b9cfd00058572243044c7b6632726a89
+size 4994944296

model-00014-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:751e8bed8d630c8c46d1fa7466c34b8703f93d57acab8ca76f32d33e447d6743
+size 4994944392

model-00015-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e20545f0a96cf537b1a222a8344c74f750ac1605efcfcc3aed39fc274d3d5d51
+size 4994944648

model-00016-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cc41909924a6d1866940342fe060909585c9d4e1d603d877ef45b3af9f6bfc70
+size 4999181952

model-00017-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:df9d63f028e6c4980df72e791eaddf29f3e492fed579fab7c4cea37965a64895
+size 4994944288

model-00018-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8b4bd5df0ccf2a34f10b3a8b6e323eb3e84f3443f830ddbc4de58c04b72d9e02
+size 4994944296

model-00019-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:918f36bdd51fdfb7545866ef568bb22f6806b7303fa4dea0b90b93e5020ab801
+size 4994944536

model-00020-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b4a9b5ce2f5051eb15612a0bee4d7af03cd3561112bee2b99f677c51b47dd84f
+size 4767862480

model-00021-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f2539882f19a491fbb8d35430a97b56519de350ee32a8db415321ed8848e7255
+size 5000091936

model-00022-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:332bd162e119c8d2976fb6f8d860e11b905bfcfdf269896e8866b57fba062769
+size 4994944296

model-00023-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1a84c3bf5b86d5d1ab112c5391293426f65ccdede9d81d89e94b752cb6bc7c36
+size 4994944392

model-00024-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ea4155bff895f55d944cc2932fe60bbe5d0063a07563bbacddb18c454fbfed6f
+size 4994944648

model-00025-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:494330e312c06cc83970f600dd614afb18c31bbea932defe0418a01dd9af3fcd
+size 4999181952

model-00026-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:08fef0bf35c5a07ba9be720f1a769c84e374877edf307883c27a164e9ec6127c
+size 4994944288

model-00027-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:434d8f6357828f5b90c50b39e61f66e71664217151f17bce8d72330635c384bd
+size 4994944296

model-00028-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ab90d0ee909ea8ae7764c2a4dd35d707c9ec1e00e017fc3f393f90d7c2328b31
+size 4994944536

model-00029-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7103c2ec83a55fda632765a375cd5c6c0a92891b6b4d47f45f5573e61a7d5380
+size 4767862480

model-00030-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:583998027f5955b5b9fd4f056e59940fd63e67fb9f8fb11f60234e6e012a4a2d
+size 5000091936

model-00031-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0e2ce43497a1405b8292442d123a6bc889127cd89d1932975796b9de2582cdef
+size 4994944296

model-00032-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7c6d7badd691827eb515b22ad8eb16fe3935f63baae82b18c8d749bfa6768a29
+size 4994944392

model-00033-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:da32ce9859d1d8bb82b060a775578f1ade1efdc12b293391e90d9787037e56b6
+size 4994944648

model-00034-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:49150c80ca6d531a51bbb6c0053e932cfb51c2a4a14d1e2b375eadd71d9f82e3
+size 4999181952

model-00035-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5152ae536963a9eb5c270becfd78c8c8157b5cc703a9fa1410af230243a9e328
+size 4994944288

model-00036-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:78c64237fecda8e591c2fc01c082dd92dc64d2bda7c5f5d3eea270322a875ba2
+size 4994944296

model-00037-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a3853e835dc9a6493a5c9a01f06d2a926afb38cc923b4f140de6d01c264f0728
+size 4994944536

model-00038-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:98264e9c269c579ba6ab5839327942158eb4cd1f6417db3093e5238e92a20c99
+size 4767862480

model-00039-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f9739f520fb31bf6f7f3474495e1bdca2bb27571206c49fe40837ac86ecbc12
+size 5000091936

model-00040-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7a63745e6393554891a5ddcecf73814fb785b4baff30c270224cbc3b083a423c
+size 4994944296

model-00041-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4c437acf47efd1616a4685feaafba61d9eeb0afa539967db98f497ca42e7620c
+size 4994944392

model-00042-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dca3b4c55b4b85f9ca7370c8b678d72a0eacb320d3656af13f56a40b3b44b9cd
+size 4994944648

model-00043-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ec53416f5c28567f50acba9ec99ecdc9c36efa67bdc47f66450b52e61e2570ae
+size 4999181952

model-00044-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:042dbfb110634ef17aacad948f22fac2da1d9b6028e2a26e9228114629053156
+size 4994944288

model-00045-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4cbaa9c03aa658d42d25a2495756b78786fd789b41a3288d5a06a96bb02cc202
+size 4994944296

model-00046-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f60ba5a10e0ec88dbf62e0a27a1f0316939c655bb86718c958d65e229c7cced3
+size 4994944536

model-00047-of-00076.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e7496e5a1f159705c405a5ddee082dbbbd99485138c0d44d3389d58455b28d32
+size 4767862480