vllm_omni.worker_v2 ¶
Modules:
| Name | Description |
|---|---|
delivery | |
first_audio_sender | Deliver in-stage first audio as soon as its device copy completes. |
model_states | Omni-aware |
native_output_worker | Materialize native producer output independently of scheduler consumption. |
omni_ar_model_runner | OmniARModelRunner — autoregressive stage runner on MR V2. |
omni_data_plane | |
omni_generation_model_runner | OmniGenerationModelRunner — non-autoregressive stage runner on MR V2. |
omni_model_runner | Omni v2 GPU model runner hooks. |
output_snapshot | Owned, dtype-grouped snapshots for asynchronous runner outputs. |