vllm_omni.worker ¶

Modules:

Name	Description
`base`	Base worker class for vLLM-Omni with device-level GPU memory profiling.
`gpu_ar_model_runner`	AR GPU Model Runner for vLLM-Omni.
`gpu_ar_worker`
`gpu_generation_model_runner`	Code2Wav GPU Model Runner for vLLM-Omni.
`gpu_generation_worker`
`gpu_memory_utils`	NVML-based per-process GPU memory utilities.
`gpu_model_runner`
`memory_utils`	GPU memory utilities for vLLM Omni workers.
`mixins`
`omni_connector_model_runner_mixin`	Public model-runner interface for Omni connector transport.
`omni_connector_validation`	Validation for Omni connector support on selected LLM workers.
`output`	Worker output helpers.
`payload_span`	Helpers for explicit thinker decode span metadata.
`runner_assisted_metadata`
`sampling_utils`	Sampling helpers shared by the GPU and NPU AR model runners.
`sparse_audio`	Sparse-audio output protocol: marker classification and payload routing.