vllm_omni.diffusion ¶

Modules:

Name	Description
`attention`
`cache`	Cache module for diffusion model inference acceleration.
`compile`
`config`	Global diffusion config — mirrors vLLM's set_current_vllm_config pattern.
`data`
`diffusion_engine`
`distributed`	Distributed utilities for vLLM-Omni diffusion models.
`envs`
`executor`
`forward_context`
`hooks`	Hook mechanism for model forward interception.
`inline_stage_diffusion_client`	Inline Stage Diffusion Client for vLLM-Omni multi-stage runtime.
`ipc`	IPC utilities for transferring large tensors via POSIX shared memory.
`layers`
`lora`
`model_loader`
`model_metadata`
`models`	Diffusion model implementations.
`offloader`
`postprocess`	Diffusion post-processing helpers.
`profiler`
`registry`
`request`
`sched`
`stage_diffusion_client`	Stage Diffusion Client for vLLM-Omni multi-stage runtime.
`stage_diffusion_proc`	Subprocess entry point for the diffusion engine.
`utils`
`worker`	Worker classes for diffusion models.