Skip to content

vllm_omni.diffusion.output_formatter

DiffusionPostprocessOutput dataclass

action_payload class-attribute instance-attribute 

action_payload: Any | None = None

audio_payload class-attribute instance-attribute 

audio_payload: Any | None = None

audio_sample_rate class-attribute instance-attribute 

audio_sample_rate: int | None = None

custom_output instance-attribute 

custom_output: dict[str, Any]

fps class-attribute instance-attribute 

fps: float | None = None

has_video_payload class-attribute instance-attribute 

has_video_payload: bool = False

outputs instance-attribute 

outputs: Any

DiffusionStepTimings dataclass

exec_time_s instance-attribute 

exec_time_s: float

postprocess_time_s instance-attribute 

postprocess_time_s: float

preprocess_time_s instance-attribute 

preprocess_time_s: float

total_time_ms instance-attribute 

total_time_ms: float

format_diffusion_outputs 

format_diffusion_outputs(
    *,
    request: OmniDiffusionRequest,
    od_config: OmniDiffusionConfig,
    diffusion_output: DiffusionOutput,
    output_data: Any,
    postprocess_output: DiffusionPostprocessOutput,
    timings: DiffusionStepTimings,
) -> list[OmniRequestOutput]

Convert a finished diffusion model output into API-facing outputs.

format_empty_diffusion_outputs 

format_empty_diffusion_outputs(
    request: OmniDiffusionRequest,
) -> list[OmniRequestOutput]

normalize_diffusion_postprocess_output 

normalize_diffusion_postprocess_output(
    outputs: Any, custom_output: dict[str, Any]
) -> DiffusionPostprocessOutput

Normalize the legacy postprocess dict shape used by diffusion models.

The returned envelope owns a shallow merged copy of custom_output so postprocess values keep legacy override precedence without mutating the caller-owned dict.