[Quantization] Add field to skip unquantized modules for GPTQ config (#25455)

Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn>

vllm/model_executor/models/qwen3_moe.py

@@ -46,9 +46,6 @@ from vllm.model_executor.layers.linear import (MergedColumnParallelLinear,
                                                RowParallelLinear)
 from vllm.model_executor.layers.logits_processor import LogitsProcessor
 from vllm.model_executor.layers.quantization import QuantizationConfig
-from vllm.model_executor.layers.quantization.gptq import GPTQConfig
-from vllm.model_executor.layers.quantization.gptq_marlin import (
-    GPTQMarlinConfig)
 from vllm.model_executor.layers.rotary_embedding import get_rope
 from vllm.model_executor.layers.vocab_parallel_embedding import (
     ParallelLMHead, VocabParallelEmbedding)
@@ -149,24 +146,11 @@ class Qwen3MoeSparseMoeBlock(nn.Module):
                                 enable_eplb=self.enable_eplb,
                                 num_redundant_experts=self.n_redundant_experts)
 
-        self.gate = ReplicatedLinear(
-            config.hidden_size,
-            config.num_experts,
-            bias=False,
-            quant_config=self._maybe_ignore_quant_config(quant_config),
-            prefix=f"{prefix}.gate")
-
-    def _maybe_ignore_quant_config(self, quant_config: QuantizationConfig):
-        # GPTQ configs do not have a list of ignored modules, however AutoGPTQ
-        # seems to avoid gate quantization while AutoRound does.
-        # See: https://huggingface.co/Qwen/Qwen3-30B-A3B-GPTQ-Int4,
-        # and https://huggingface.co/jart25/Qwen3-Coder-30B-A3B-Instruct-Int4-gptq
-        if isinstance(
-                quant_config,
-            (GPTQConfig,
-             GPTQMarlinConfig)) and not quant_config.autoround_version:
-            return None
-        return quant_config
+        self.gate = ReplicatedLinear(config.hidden_size,
+                                     config.num_experts,
+                                     bias=False,
+                                     quant_config=quant_config,
+                                     prefix=f"{prefix}.gate")
 
     def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
         assert hidden_states.dim(
@@ -699,4 +683,4 @@ class Qwen3MoeForCausalLM(nn.Module, SupportsPP, SupportsLoRA,
         return loader.load_weights(weights)
 
     def get_expert_mapping(self) -> list[tuple[str, str, int, str]]:
-        return self.model.get_expert_mapping()
+        return self.model.get_expert_mapping()