Fix: AWQ Marlin get_quant_method does not recognize "modules_to_not_convert" (#21888)

Signed-off-by: JunHowie <[email protected]>
Co-authored-by: JunHowie <[email protected]>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>

Author: 3 people

vllm/model_executor/layers/quantization/awq_marlin.py

@@ -10,7 +10,8 @@
 from vllm import _custom_ops as ops
 from vllm.logger import init_logger
 from vllm.model_executor.layers.fused_moe.layer import (
-    FusedMoE, FusedMoEMethodBase, FusedMoeWeightScaleSupported)
+    FusedMoE, FusedMoEMethodBase, FusedMoeWeightScaleSupported,
+    UnquantizedFusedMoEMethod)
 from vllm.model_executor.layers.linear import (LinearBase, LinearMethodBase,
                                                UnquantizedLinearMethod,
                                                set_weight_attrs)
@@ -141,6 +142,9 @@ def get_quant_method(self, layer: torch.nn.Module,
         elif isinstance(layer, FusedMoE):
             from vllm.model_executor.layers.quantization.moe_wna16 import (
                 MoeWNA16Config)
+            if is_layer_skipped_awq(
+                    prefix, getattr(self, "modules_to_not_convert", [])):
+                return UnquantizedFusedMoEMethod(layer.moe_config)
             if not check_moe_marlin_supports_layer(layer, self.group_size):
                 logger.warning_once(
                     f"Layer '{prefix}' is not supported by AWQMoeMarlin. "
@@ -520,4 +524,4 @@ def apply(
             expert_map=expert_map,
             w1_zeros=layer.w13_qzeros,
             w2_zeros=layer.w2_qzeros,
-            workspace=layer.workspace)
+            workspace=layer.workspace)