Skip to content

vllm_omni.model_executor.models ¶

Modules:

Name	Description
`audex`
`audio8_tts`	Audio8 TTS Preview (DualAR, 44.1 kHz codec) model implementation.
`auk`
`aura_omni`
`bagel`
`breeze_tts_2`
`common`
`cosmos3`
`cosyvoice3`
`covo_audio`
`dreamzero`
`fish_speech`
`gepard`
`glm_image`
`glm_tts`
`gr00t`
`higgs_audio_v2`
`higgs_audio_v3`
`hunyuan_image3`
`hunyuan_video`
`indextts2`
`joyai_vl_interaction`	JoyAI-VL-Interaction native pipeline.
`lance`
`lingbot_world`
`mammoth_moda2`
`mimo_audio`
`ming_flash_omni`
`ming_image`
`ming_tts`
`minicpmo_4_5`
`minimax_h3`	MiniMax H3 model components.
`minimax_music3`	MiniMax Music 3 text-to-music.
`model_local_kv`	Declaration protocol for attention KV kept outside the paged manager.
`moss_tts`
`moss_tts_nano`
`nemotron_voicechat`	NVIDIA NemotronLabs VoiceChat-11B: offline speech-to-speech 3-stage pipeline.
`omnivoice`
`output_templates`
`personaplex`	vllm-omni integration for PersonaPlex (a Moshi finetune, full-duplex S2S).
`qwen2_5_omni`
`qwen3_omni`
`qwen3_tts`
`registry`
`step_audio2`
`utils`
`voxcpm2`
`voxtral_tts`
`wan2_2`
`whisper_utils`

OmniModelRegistry `module-attribute` ¶

OmniModelRegistry = _ModelRegistry(
    {
        **{
            model_arch: _LazyRegisteredModel(
                module_name=_resolve_module_name(
                    mod_relname
                ),
                class_name=cls_name,
            )
            for model_arch, (
                mod_relname,
                cls_name,
            ) in _VLLM_MODELS.items()
        },
        **{
            model_arch: _LazyRegisteredModel(
                module_name=f"vllm_omni.model_executor.models.{mod_folder}.{mod_relname}",
                class_name=cls_name,
            )
            for model_arch, (
                mod_folder,
                mod_relname,
                cls_name,
            ) in _OMNI_MODELS.items()
        },
    }
)