Skip to content

vllm_omni.model_executor.models.omnivoice.pipeline ¶

OmniVoice pipeline topology (frozen).

Single-stage diffusion model for text-to-speech. Stage 0: DiT — text → iterative unmasking → codebook tokens → DAC decode → 24kHz audio

OMNIVOICE_PIPELINE `module-attribute` ¶

OMNIVOICE_PIPELINE = PipelineConfig(
    model_type="omnivoice",
    model_arch="OmniVoicePipeline",
    stages=(
        StagePipelineConfig(
            stage_id=0,
            model_stage="dit",
            execution_type=StageExecutionType.DIFFUSION,
            input_sources=(),
            final_output=True,
            final_output_type="audio",
            owns_tokenizer=False,
            requires_multimodal_data=False,
            engine_output_type="audio",
        ),
    ),
)