Skip to content

vllm_omni.model_executor.models.mammoth_moda2.pipeline

MammothModa2 pipeline topology (frozen).

Stage 0: AR — multimodal understanding + latent generation Stage 1: DiT — latent → image

For text/image understanding tasks (text output only), use MAMMOTH_MODA2_AR_PIPELINE.

MAMMOTH_MODA2_AR_PIPELINE module-attribute 

MAMMOTH_MODA2_AR_PIPELINE = PipelineConfig(
    model_type="mammoth_moda2_ar",
    model_arch="MammothModa2ForConditionalGeneration",
    stages=(
        StagePipelineConfig(
            stage_id=0,
            model_stage="ar",
            execution_type=StageExecutionType.LLM_AR,
            input_sources=(),
            final_output=True,
            final_output_type="text",
            owns_tokenizer=True,
            requires_multimodal_data=True,
            engine_output_type="text",
        ),
    ),
)

MAMMOTH_MODA2_PIPELINE module-attribute 

MAMMOTH_MODA2_PIPELINE = PipelineConfig(
    model_type="mammoth_moda2",
    model_arch="MammothModa2ForConditionalGeneration",
    hf_architectures=(
        "Mammothmoda2Model",
        "MammothModa2ForConditionalGeneration",
    ),
    stages=(
        StagePipelineConfig(
            stage_id=0,
            model_stage="ar",
            execution_type=StageExecutionType.LLM_AR,
            input_sources=(),
            final_output=False,
            owns_tokenizer=True,
            requires_multimodal_data=True,
            engine_output_type="latent",
        ),
        StagePipelineConfig(
            stage_id=1,
            model_stage="dit",
            execution_type=StageExecutionType.LLM_GENERATION,
            input_sources=(0,),
            final_output=True,
            final_output_type="image",
            owns_tokenizer=False,
            requires_multimodal_data=False,
            engine_output_type="image",
            custom_process_input_func=f"{_PROC}.ar2dit",
        ),
    ),
)