Merge branch 'v0.11.0-dev-wm' into 'v0.11.0-dev'

[fix]解决GLM-w8a8+mtp启动失败问题

See merge request dcutoolkit/deeplearing/vllm!435

vllm/config/model.py

@@ -928,6 +928,7 @@ class ModelConfig:
                 # imports during override detection (e.g., MXFP4 imports Triton)
                 "mxfp4",
                 "slimquant_w4a8_marlin",
+                "slimquant_marlin",
                 "slimquant_compressed_tensors_marlin",
             ]
             quantization_methods = [

vllm/model_executor/layers/quantization/__init__.py

@@ -38,6 +38,7 @@ QuantizationMethods = Literal[
     "blockwise_int8",
     "slimquant_w4a8",
     "slimquant_w4a8_marlin",
+    "slimquant_marlin",
     "slimquant_compressed_tensors_marlin",
 ]
 QUANTIZATION_METHODS: list[str] = list(get_args(QuantizationMethods))
@@ -154,6 +155,7 @@ def get_quantization_config(quantization: str) -> type[QuantizationConfig]:
         "blockwise_int8": BlockInt8Config,
         "slimquant_w4a8":SlimQuantW4A8Int8Config,
         "slimquant_w4a8_marlin":SlimQuantW4A8Int8MarlinConfig,
+        "slimquant_marlin":SlimQuantCompressedTensorsMarlinConfig,
         "slimquant_compressed_tensors_marlin":SlimQuantCompressedTensorsMarlinConfig,
     }
     # Update the `method_to_config` with customized quantization methods.