BLOG ILLUSTRATION / EXECUTION STORY

One speech request through three model stages

Two source-derived traces follow one streaming Qwen3-Omni request that returns text and speech. vLLM-Omni runs each stage as a vLLM engine. SGLang-Omni runs each stage as a node in a coordinator graph. Select either trace on this page.

One speech request through three model stages A frontend process with a controller submits a Qwen3-Omni request to the Thinker stage. The Thinker streams to the Talker, the Talker streams codec chunks to Code2Wav, and Code2Wav returns audio chunks to the frontend. Text returns either directly from the Thinker or through a separate text decode stage. A dashed control path pre-submits a placeholder request to the Talker. Frontend controller Thinker text model Talker codec model Code2Wav vocoder Text decode detokenizer request text tokens text delta token IDs pre-submits placeholder audio chunk Solid: data or output Dashed: control
Author-created trace from vLLM-Omni v0.32.0 and SGLang-Omni v0.2.0 default Qwen3-Omni configurations. Encoder and preprocessing stages are not drawn. Playback time is not measured latency. Open full-size SVG