[Bugfix] fix w8a8_int8 load issue (#8308)

Co-authored-by: ronnie_zheng <zl19940307@163.com>

python/sglang/srt/layers/quantization/w8a8_int8.py

@@ -231,7 +231,10 @@ class W8A8Int8Config(QuantizationConfig):
 
     @classmethod
     def get_config_filenames(cls) -> List[str]:
-        return []
+        filenames = []
+        if _is_npu:
+            filenames.append("quant_model_description.json")
+        return filenames
 
     @classmethod
     def from_config(cls, config: Dict[str, Any]) -> W8A8Int8Config:

python/sglang/srt/model_loader/weight_utils.py

@@ -229,6 +229,8 @@ def get_quant_config(
                     f"Unsupported quantization config"
                     f" found for {model_config.quantization} in {f}."
                 )
+        elif model_config.quantization == "w8a8_int8":
+            config["packed_modules_mapping"] = packed_modules_mapping
 
     return quant_cls.from_config(config)