[diffusion] chore: improve the output_path config and enable the server to return inference duration (#16965)

Co-authored-by: Xiaoyu Zhang <35585791+BBuf@users.noreply.github.com>

python/sglang/multimodal_gen/configs/sample/sampling_params.py

@@ -318,6 +318,12 @@ class SamplingParams:
 
         self.data_type = server_args.pipeline_config.task_type.data_type()
 
+        if server_args.output_path is not None:
+            self.output_path = server_args.output_path
+            logger.debug(
+                f"Overriding output_path with server configuration: {self.output_path}"
+            )
+
         # Process negative prompt
         if self.negative_prompt is not None and not self.negative_prompt.isspace():
             # avoid stripping default negative prompt: ' ' for qwen-image
@@ -533,12 +539,6 @@ class SamplingParams:
             default=SamplingParams.prompt_path,
             help="Path to a text file containing the prompt",
         )
-        parser.add_argument(
-            "--output-path",
-            type=str,
-            default=SamplingParams.output_path,
-            help="Path to save the generated image/video",
-        )
         parser.add_argument(
             "--output-file-name",
             type=str,

python/sglang/multimodal_gen/runtime/entrypoints/openai/protocol.py

@@ -17,6 +17,7 @@ class ImageResponse(BaseModel):
     created: int = Field(default_factory=lambda: int(time.time()))
     data: List[ImageResponseData]
     peak_memory_mb: Optional[float] = None
+    inference_time_s: Optional[float] = None
 
 
 class ImageGenerationsRequest(BaseModel):
@@ -60,6 +61,7 @@ class VideoResponse(BaseModel):
     error: Optional[Dict[str, Any]] = None
     file_path: Optional[str] = None
     peak_memory_mb: Optional[float] = None
+    inference_time_s: Optional[float] = None
 
 
 class VideoGenerationsRequest(BaseModel):

python/sglang/multimodal_gen/runtime/entrypoints/openai/utils.py

@@ -288,6 +288,9 @@ def add_common_data_to_response(
     if result.peak_memory_mb and result.peak_memory_mb > 0:
         response["peak_memory_mb"] = result.peak_memory_mb
 
+    if result.timings and result.timings.total_duration_s > 0:
+        response["inference_time_s"] = result.timings.total_duration_s
+
     response["id"] = request_id
 
     return response

python/sglang/multimodal_gen/runtime/server_args.py

@@ -328,6 +328,8 @@ class ServerArgs:
 
     scheduler_port: int = 5555
 
+    output_path: str | None = None
+
     # Prompt text file for batch processing
     prompt_file_path: str | None = None
 
@@ -682,6 +684,12 @@ class ServerArgs:
             default=ServerArgs.webui_port,
             help="Whether to use webui for better display",
         )
+        parser.add_argument(
+            "--output-path",
+            type=str,
+            default=ServerArgs.output_path,
+            help="Directory path to save generated images/videos",
+        )
 
         # LoRA
         parser.add_argument(