Skip to content

vllm_omni.diffusion.models.minimax_h3.chunked_decode ¶

Frames-only callback coordination for MiniMax-H3 VAE decoding.

MiniMaxH3VideoChunkCallback `module-attribute` ¶

MiniMaxH3VideoChunkCallback = Callable[[torch.Tensor], None]

decode_h3_chunks ¶

decode_h3_chunks(
    host,
    latent: Tensor,
    callback: MiniMaxH3VideoChunkCallback | None,
    *,
    group: ProcessGroup | None,
) -> Tensor

Run the local temporal loop, synchronizing callback errors across ranks.