[3/N][Refactor][Quantization]remove packed_modules_mapping from models

Signed-off-by: 22dimensions <waitingwind@foxmail.com>

Author: 22dimensions

vllm_ascend/models/deepseek_mtp.py

@@ -180,14 +180,6 @@ def compute_logits(
 
 
 class CustomDeepSeekMTP(DeepSeekMTP):
-    # NOTE 1.The quantized MTP layer of deepseek on the NPU is not quantized;
-    # NOTE 2.The description file generated by the current msmodelslim tool does not have
-    # MTP layer info. Please manually add it and set the value to FLOAT.
-    packed_modules_mapping = {
-        "gate_up_proj": ["gate_proj", "up_proj"],
-        "experts":
-        ["experts.0.gate_proj", "experts.0.up_proj", "experts.0.down_proj"]
-    }
 
     def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
         nn.Module.__init__(self)

vllm_ascend/models/qwen2_5_vl.py

@@ -491,17 +491,6 @@ def forward(
     dummy_inputs=Qwen2_5_VLDummyInputsBuilder)
 class AscendQwen2_5_VLForConditionalGeneration(
         Qwen2_5_VLForConditionalGeneration):
-    packed_modules_mapping = {
-        "qkv_proj": [
-            "q_proj",
-            "k_proj",
-            "v_proj",
-        ],
-        "gate_up_proj": [
-            "gate_proj",
-            "up_proj",
-        ],
-    }
 
     def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
         super().__init__(vllm_config=vllm_config, prefix=prefix)

vllm_ascend/models/qwen3_moe.py

@@ -318,19 +318,6 @@ def forward(
 
 
 class CustomQwen3MoeForCausalLM(Qwen3MoeForCausalLM):
-    packed_modules_mapping = {
-        "qkv_proj": [
-            "q_proj",
-            "k_proj",
-            "v_proj",
-        ],
-        "gate_up_proj": [
-            "gate_proj",
-            "up_proj",
-        ],
-        "experts":
-        ["experts.0.gate_proj", "experts.0.up_proj", "experts.0.down_proj"],
-    }
 
     def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
         nn.Module.__init__(self)

vllm_ascend/models/qwen3_next.py

@@ -1166,15 +1166,6 @@ def load_weights(self, weights: Iterable[tuple[str,
 
 class Qwen3NextForCausalLM(nn.Module, HasInnerState, SupportsLoRA, SupportsPP,
                            MixtureOfExperts, IsHybrid):
-    packed_modules_mapping = {
-        "qkv_proj": [
-            "q_proj",
-            "k_proj",
-            "v_proj",
-        ],
-        "gate_up_proj": ["gate_proj", "up_proj"],
-        "in_proj": ["in_proj_qkvz", "in_proj_ba"],
-    }
 
     def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
         config = vllm_config.model_config.hf_config

vllm_ascend/quantization/quant_config.py

@@ -19,6 +19,7 @@
 from typing import Any, Callable, Dict, List, Mapping, Optional
 
 import torch
+from vllm.config import get_current_vllm_config
 from vllm.distributed import get_tensor_model_parallel_rank
 from vllm.model_executor.layers.fused_moe import (FusedMoE, FusedMoEMethodBase,
                                                   FusedMoeWeightScaleSupported)
@@ -55,6 +56,11 @@ class AscendQuantConfig(QuantizationConfig):
     def __init__(self, quant_config: Dict[str, Any]):
         super().__init__()
         self.quant_description = quant_config
+        vllm_config = get_current_vllm_config()
+        model_type = vllm_config.model_config.hf_config.model_type
+        if model_type in packed_modules_model_mapping:
+            self.packed_modules_mapping = packed_modules_model_mapping[
+                model_type]
 
     def __repr__(self) -> str:
         return "AscendQuantConfig:\n" + super().__repr__()
@@ -153,6 +159,42 @@ def get_scaled_act_names(self) -> List[str]:
         return []
 
 
+packed_modules_model_mapping = {
+    "qwen3_moe": {
+        "qkv_proj": [
+            "q_proj",
+            "k_proj",
+            "v_proj",
+        ],
+        "gate_up_proj": [
+            "gate_proj",
+            "up_proj",
+        ],
+        "experts":
+        ["experts.0.gate_proj", "experts.0.up_proj", "experts.0.down_proj"],
+    },
+    "deepseek_v2": {
+        "gate_up_proj": ["gate_proj", "up_proj"],
+        "experts":
+        ["experts.0.gate_proj", "experts.0.up_proj", "experts.0.down_proj"]
+    },
+    "deepseek_mtp": {
+        "gate_up_proj": ["gate_proj", "up_proj"],
+        "experts":
+        ["experts.0.gate_proj", "experts.0.up_proj", "experts.0.down_proj"]
+    },
+    "qwen3_next": {
+        "qkv_proj": [
+            "q_proj",
+            "k_proj",
+            "v_proj",
+        ],
+        "gate_up_proj": ["gate_proj", "up_proj"],
+        "in_proj": ["in_proj_qkvz", "in_proj_ba"],
+    }
+}
+
+
 class AscendLinearMethod(LinearMethodBase):
     """Linear method for Ascend quantization.