Client-side libraries for vLLM-Omni serving APIs.
vllm_omni.clients.duplex speaks to a running vLLM-Omni server over the network; vllm_omni.clients.inline_duplex drives an in-process DuplexOmni handed in by the caller with the same client API. Modules here import no server runtime code at module load, and server runtime code (vllm_omni.engine, vllm_omni.entrypoints, vllm_omni.model_executor) must never import from here.
Modules:
| Name | Description |
|---|---|
duplex | Public asynchronous client for the vLLM-Omni full-duplex Realtime API. |
inline_duplex | In-process duplex client: the :class: |
minicpmo_4_5 | Client-side duplex session preset for MiniCPM-o 4.5 native duplex. |
nemotron_voicechat | Client-side duplex session preset for Nemotron VoiceChat. |
personaplex | Client-side duplex session preset for PersonaPlex duplex serving. |
utils | Client-local helpers for duplex JSON metric blocks. |