vllm_omni.diffusion.io_support ¶

get_diffusion_output_type ¶

get_diffusion_output_type(
    model_class_name: str | None,
) -> str

Return the declared final modality for a diffusion model.

get_dummy_run_num_frames(
    model_class_name: str, supports_audio_input: bool
) -> int

Get num_frames for the dummy warmup run. Returns 0 to skip warmup.

get_dummy_run_num_image_inputs(
    model_class_name: str,
) -> int

Return the maximum advertised image-input count for profiling.

image_color_format(model_class_name: str) -> str

supports_audio_output(model_class_name: str) -> bool

supports_multimodal_input(
    od_config: OmniDiffusionConfig,
) -> tuple[bool, bool]