⚡️ Speed up method MoeWNA16Config.from_config by 28%

python/sglang/srt/layers/quantization/base_config.py

@@ -117,7 +117,7 @@ class QuantizationConfig(ABC):
     def __init__(self):
         super().__init__()
         # mapping is updated by models as they initialize
-        self.packed_modules_mapping: Dict[str, List[str]] = dict()
+        self.packed_modules_mapping: Dict[str, List[str]] = {}
 
     @abstractmethod
     def get_name(self) -> str:
@@ -175,10 +175,10 @@ def get_from_keys(config: Dict[str, Any], keys: List[str]) -> Any:
     @staticmethod
     def get_from_keys_or(config: Dict[str, Any], keys: List[str], default: Any) -> Any:
         """Get a optional value from the model's quantization config."""
-        try:
-            return QuantizationConfig.get_from_keys(config, keys)
-        except ValueError:
-            return default
+        for key in keys:
+            if key in config:
+                return config[key]
+        return default
 
     @abstractmethod
     def get_quant_method(

python/sglang/srt/layers/quantization/moe_wna16.py

@@ -104,10 +104,7 @@ def __init__(
         else:
             raise ValueError("moe_wna16 only support gptq and awq.")
 
-        if modules_to_not_convert is None:
-            self.modules_to_not_convert = []
-        else:
-            self.modules_to_not_convert = modules_to_not_convert
+        self.modules_to_not_convert = modules_to_not_convert or []
 
     @classmethod
     def get_name(cls) -> str: