Skip to content

vllm.models.muse_spark

Muse Spark HF model entry point.

Modules:

  • amd –

    AMD ROCm implementation of MuseSpark1x.

  • common –

    Platform-independent MuseSpark preprocessing.

  • nvidia –

    NVIDIA-specific MuseSpark model implementations.

Classes:

MuseSparkForConditionalGeneration

Bases: MuseSpark1xOmniMixin, MuseSpark1xForCausalLM, SupportsMultiModal

MuseSpark1x Omni: ROCm text stack plus NaViT vision and waveform audio.

Source code in vllm/models/muse_spark/amd/multimodal.py 
15
16
17
18
19
20
21
@register_omni_processor
class MuseSparkForConditionalGeneration(
    MuseSpark1xOmniMixin, MuseSpark1xForCausalLM, SupportsMultiModal
):
    """MuseSpark1x Omni: ROCm text stack plus NaViT vision and waveform audio."""

    hf_to_vllm_mapper = MuseSpark1xForCausalLM.hf_to_vllm_mapper | OMNI_WEIGHTS_MAPPER