Enable native ModelOpt quantization support (3/3) (#10154)

Signed-off-by: Zhiyu Cheng <zhiyuc@nvidia.com>

python/sglang/srt/server_args.py

@@ -83,6 +83,7 @@ QUANTIZATION_CHOICES = [
     "bitsandbytes",
     "gguf",
     "modelopt",
+    "modelopt_fp8",
     "modelopt_fp4",
     "petit_nvfp4",
     "w8a8_int8",
@@ -192,6 +193,8 @@ class ServerArgs:
     modelopt_quant: Optional[Union[str, Dict]] = None
     modelopt_checkpoint_restore_path: Optional[str] = None
     modelopt_checkpoint_save_path: Optional[str] = None
+    modelopt_export_path: Optional[str] = None
+    quantize_and_serve: bool = False
     context_length: Optional[int] = None
     is_embedding: bool = False
     enable_multimodal: Optional[bool] = None
@@ -1743,6 +1746,22 @@ class ServerArgs:
             help="Path to save the ModelOpt quantized checkpoint after quantization. "
             "This allows reusing the quantized model in future runs.",
         )
+        parser.add_argument(
+            "--modelopt-export-path",
+            type=str,
+            default=ServerArgs.modelopt_export_path,
+            help="Path to export the quantized model in HuggingFace format after ModelOpt quantization. "
+            "The exported model can then be used directly with SGLang for inference. "
+            "If not provided, the model will not be exported.",
+        )
+        parser.add_argument(
+            "--quantize-and-serve",
+            action="store_true",
+            default=ServerArgs.quantize_and_serve,
+            help="Quantize the model with ModelOpt and immediately serve it without exporting. "
+            "This is useful for development and prototyping. For production, it's recommended "
+            "to use separate quantization and deployment steps.",
+        )
         parser.add_argument(
             "--kv-cache-dtype",
             type=str,