[Bugfix] fix moe_wna16 get_quant_method (#12648)

Fix https://github.com/vllm-project/vllm/issues/12647
The `get_quant_method` of `moe_wna16` always return moe method,
GPTQ-based linear method or AWQ-based linear method, even when the
target module is attention layer.


baeded2569/vllm/attention/layer.py (L86-L92)

Signed-off-by: Jinzhen Lin <linjinzhen@hotmail.com>

vllm/model_executor/layers/quantization/moe_wna16.py

@@ -6,16 +6,13 @@ from vllm.distributed import get_tensor_model_parallel_rank, get_tp_group
 from vllm.model_executor.layers.fused_moe.layer import (
     FusedMoE, FusedMoEMethodBase, FusedMoeWeightScaleSupported)
 from vllm.model_executor.layers.linear import UnquantizedLinearMethod
-from vllm.model_executor.layers.quantization.awq import (AWQConfig,
+from vllm.model_executor.layers.quantization.awq import AWQConfig
-                                                         AWQLinearMethod)
+from vllm.model_executor.layers.quantization.awq_marlin import AWQMarlinConfig
-from vllm.model_executor.layers.quantization.awq_marlin import (
-    AWQMarlinConfig, AWQMarlinLinearMethod)
 from vllm.model_executor.layers.quantization.base_config import (
     QuantizationConfig, QuantizeMethodBase)
-from vllm.model_executor.layers.quantization.gptq import (GPTQConfig,
+from vllm.model_executor.layers.quantization.gptq import GPTQConfig
-                                                          GPTQLinearMethod)
 from vllm.model_executor.layers.quantization.gptq_marlin import (
-    GPTQMarlinConfig, GPTQMarlinLinearMethod)
+    GPTQMarlinConfig)
 from vllm.model_executor.utils import set_weight_attrs
 from vllm.platforms import current_platform
 
@@ -131,18 +128,18 @@ class MoeWNA16Config(QuantizationConfig):
         else:
             if self.linear_quant_method == "gptq":
                 if self.use_marlin:
-                    return GPTQMarlinLinearMethod(
+                    return GPTQMarlinConfig.from_config(
-                        GPTQMarlinConfig.from_config(self.full_config))
+                        self.full_config).get_quant_method(layer, prefix)
                 else:
-                    return GPTQLinearMethod(
+                    return GPTQConfig.from_config(
-                        GPTQConfig.from_config(self.full_config))
+                        self.full_config).get_quant_method(layer, prefix)
             elif self.linear_quant_method == "awq":
                 if self.use_marlin:
-                    return AWQMarlinLinearMethod(
+                    return AWQMarlinConfig.from_config(
-                        AWQMarlinConfig.from_config(self.full_config))
+                        self.full_config).get_quant_method(layer, prefix)
                 else:
-                    return AWQLinearMethod(
+                    return AWQConfig.from_config(
-                        AWQConfig.from_config(self.full_config))
+                        self.full_config).get_quant_method(layer, prefix)
             else:
                 raise ValueError("moe_wna16 only support gptq and awq.")