[feat] Support tp mode for DeepSeek-R1-W4AFP8 (#8118)

Co-authored-by: yuhyao <827623970@qq.com>

python/sglang/srt/configs/model_config.py

@@ -405,9 +405,10 @@ class ModelConfig:
             # compressed-tensors uses a "compression_config" key
             quant_cfg = getattr(self.hf_config, "compression_config", None)
         if quant_cfg is None:
-            # check if is modelopt model -- modelopt doesn't have corresponding field
+            # check if is modelopt or mixed-precision model -- Both of them don't have corresponding field
             # in hf `config.json` but has a standalone `hf_quant_config.json` in the root directory
             # example: https://huggingface.co/nvidia/Llama-3.1-8B-Instruct-FP8/tree/main
+            # example: https://huggingface.co/Barrrrry/DeepSeek-R1-W4AFP8/tree/main
             is_local = os.path.exists(self.model_path)
             modelopt_quant_config = {"quant_method": "modelopt"}
             if not is_local: