[Entrypoints] online serving support usp args (#366)

Signed-off-by: David Chen <530634352@qq.com>

docs/user_guide/acceleration/parallelism_acceleration.md

@@ -40,6 +40,15 @@ outputs = omni.generate(prompt="A cat sitting on a windowsill", num_inference_st
 
 See `examples/offline_inference/text_to_image/text_to_image.py` for a complete working example.
 
+##### Online Serving
+
+You can enable Ulysses-SP in online serving for diffusion models via `--usp`:
+
+```bash
+# Text-to-image (requires >= 2 GPUs)
+vllm serve Qwen/Qwen-Image --omni --port 8091 --usp 2
+```
+
 ##### Benchmarks
 !!! note "Benchmark Disclaimer"
     These benchmarks are provided for **general reference only**. The configurations shown use default or common parameter settings and have not been exhaustively optimized for maximum performance. Actual performance may vary based on:

vllm_omni/entrypoints/cli/serve.py

@@ -134,9 +134,18 @@ class OmniServeCommand(CLISubcommand):
         serve_parser.add_argument(
             "--num-gpus",
             type=int,
-            default=1,
+            default=None,
             help="Number of GPUs to use for diffusion model inference.",
         )
+        serve_parser.add_argument(
+            "--usp",
+            "--ulysses-degree",
+            dest="ulysses_degree",
+            type=int,
+            default=None,
+            help="Ulysses Sequence Parallelism degree for diffusion models. "
+            "Equivalent to setting DiffusionParallelConfig.ulysses_degree.",
+        )
 
         # Cache optimization parameters
         serve_parser.add_argument(

vllm_omni/entrypoints/openai/api_server.py

@@ -40,7 +40,7 @@ from vllm.logger import init_logger
 from vllm.transformers_utils.tokenizer import MistralTokenizer
 from vllm.utils import decorate_logs
 
-from vllm_omni.diffusion.data import OmniDiffusionConfig
+from vllm_omni.diffusion.data import DiffusionParallelConfig, OmniDiffusionConfig
 from vllm_omni.diffusion.utils.hf_utils import is_diffusion_model
 from vllm_omni.entrypoints.async_diffusion import AsyncOmniDiffusion
 from vllm_omni.entrypoints.async_omni import AsyncOmni
@@ -236,6 +236,18 @@ async def build_async_diffusion(
         config_field_names = {f.name for f in fields(OmniDiffusionConfig)}
         diffusion_kwargs: dict[str, Any] = {"model": args.model}
 
+        # Diffusion parallelism configuration (e.g. `--usp 2`).
+        parallel_config_kwargs: dict[str, Any] = {}
+        for field in fields(DiffusionParallelConfig):
+            if not hasattr(args, field.name):
+                continue
+            value = getattr(args, field.name)
+            if value is None:
+                continue
+            parallel_config_kwargs[field.name] = value
+        if parallel_config_kwargs:
+            diffusion_kwargs["parallel_config"] = DiffusionParallelConfig(**parallel_config_kwargs)
+
         for field_name in config_field_names:
             if not hasattr(args, field_name):
                 continue