Skip to content

vllm_omni.entrypoints.duplex

Thin websocket serving for engine-resident duplex sessions.

The session logic lives in vllm_omni.engine.duplex and is reached through vllm_omni.entrypoints.duplex_omni.DuplexOmni; this package only speaks the OpenAI Realtime wire protocol over /v1/realtime?duplex=1.

Modules:

Name Description
audio_encoding

Audio encoding for duplex model output (entrypoint layer, injected into DuplexOmniEngine).
realtime_input

Thin OpenAI Realtime wire envelope for one websocket connection.
serving

Thin websocket serving for /v1/realtime?duplex=1.
session_attachment
warmup

Startup warmup probe for duplex /v1/realtime.
websocket

WebSocket transport helpers for the duplex Realtime handler.

OmniDuplexSessionHandler

WebSocket transport for engine-resident duplex sessions.

handle_realtime_session async 

handle_realtime_session(websocket: WebSocket) -> None