vllm_omni.diffusion.models.diffusers_adapter.pipeline_diffusers_adapter ¶

Diffusers backend adapter for vLLM-Omni.

Provides a black-box wrapper around any 🤗 Diffusers pipeline, enabling vLLM-Omni to directly serve Diffusers models with near-zero per-model code.

The adapter delegates full pipeline execution to diffusers' __call__(). It does NOT support: - CFG parallel (diffusers handles CFG via guidance_scale internally) - Sequence parallel (requires model-specific attention surgery) - TeaCache / Cache-DiT (requires hooking into transformer blocks) - Step-wise execution (continuous batching)

logger `module-attribute` ¶

logger = getLogger(__name__)

DiffusersAdapterPipeline ¶

Bases: Module, DiffusionPipelineProfilerMixin

Black-box adapter that delegates full pipeline execution to a diffusers pipeline.

Usage::

adapter = DiffusersAdapterPipeline(od_config=od_config)
adapter.load_weights()  # calls DiffusionPipeline.from_pretrained()
output = adapter.forward(req)

Step-wise execution is explicitly rejected — diffusers encapsulates the full denoising loop internally. Use native pipelines for continuous batching mode.

device `instance-attribute` ¶

device = device

od_config `instance-attribute` ¶

od_config = od_config

supports_step_execution `class-attribute` `instance-attribute` ¶

supports_step_execution: bool = False

denoise_step ¶

denoise_step(**_: Any) -> Tensor | None

forward ¶

forward(req: OmniDiffusionRequest) -> DiffusionOutput

Full delegation to diffusers pipeline.__call__().

load_weights ¶

load_weights() -> None

Load the diffusers pipeline via DiffusionPipeline.from_pretrained().

post_decode ¶

post_decode(**_: Any) -> Any

prepare_encode ¶

prepare_encode(**_: Any) -> Any

step_scheduler ¶

step_scheduler(**_: Any) -> None