vllm-omni

# vLLM-Omni — output-side multimodal serving Target: operators who serve image / video / audio / any-to-any generation models with the vLLM-Omni fork of vLLM. vllm-omni extends upstream vLLM (same CUDA/ROCm/NPU/XPU runtime, same OpenAI-compat API server) to add non-autoregressive DiT models, multi-stage pipeline execution, diffusion schedulers, CFG plumbing, and real-time streaming audio I/O — things upstream vLLM does not ship. This skill is a **reference**, not a tutorial. SKILL.md holds the mental model, quick-answer router, top pitfalls, and operator cheat sheet. The `references/` files hold endpoint catalogs, supported-model tables, stage-config grammar, and the diffusion/DiT details. Read only the reference file that matches the question. ## The one thing to know before anything else vllm-omni is **not a fork** — it layers on top of upstream vLLM, registers OmniModelConfig, and adds one CLI flag: `--omni`. Adding `--omni` to `vllm serve` routes the server through `vllm_omni.entrypoints`. As of v0.20.0 the old vLLM entrypoint-hijack / `patch.py` early-import mechanism was **removed** — the v0.20.0 release notes state "removal of the old vLLM entrypoint hijack, and runtime changes needed for the 0.20.0 integration path (#3232, #3082, #3352, #3393, #2306)". The omni runtime is now rebased onto upstream vLLM v0.20.0 (rebase PR #3232) rather than monkey-patching it. The architectural claim is to decompose any-to-any models into a **graph of disaggregated stages** (Thinke...