Fix trtllm-serve command for TensorRT-LLM ToT Main
Ying Guo
Verified signature
ed5341f
The TensorRT-LLM command in the Usage section fails on TensorRT-LLM 1.4.0rc0 before the model loads: `Error: No such option '--extra_visual_gen_options'`.
In 1.4.0rc0, `trtllm-serve` takes the visual-gen config via `--visual_gen_args`, and the Wan2.2 config is `examples/visual_gen/serve/configs/wan22.yml` (`wan.yml` does not exist there).
Changes:
- Use `--visual_gen_args ./examples/visual_gen/serve/configs/wan22.yml`
- Note that the command runs from the root of a TensorRT-LLM source checkout, since the config path is relative to it
- Link the TensorRT-LLM visual_gen serving README
The SGLang command is unchanged. The same fix for `nvidia/Wan2.2-T2V-A14B-Diffusers-NVFP4` is in https://huggingface.co/nvidia/Wan2.2-T2V-A14B-Diffusers-NVFP4/discussions/2.
References (TensorRT-LLM v1.4.0rc0):
- `--visual_gen_args` option: https://github.com/NVIDIA/TensorRT-LLM/blob/cc63e66daf52644f1441e117bb7ba67d20566016/tensorrt_llm/commands/serve.py#L1323
- Serving README: https://github.com/NVIDIA/Ten…
