kevin-mi/FastH3-8step-V2-overlay

FastH3 8-Step V2 Overlay

Overlay metadata repo for using FastVideo/FastVideo-FastH3-8-Step-V2 with native SGLang Diffusion MiniMax-H3 support.

This repo contains only overlay metadata and materialization logic. Source weights remain in FastVideo/FastVideo-FastH3-8-Step-V2; SGLang's registry pins a revision of this repo so re-materializing the model does not drift.

The materializer reshapes the flat native-Diffusers release into the base MiniMax-H3 partition layout SGLang's loaders consume:

• model_index.json declares the t2va-only release contract (tasks: ["t2va"], video/audio sigma shifts 10/3, and the trained eight DMD rungs [999, 874, 749, 624, 500, 375, 250, 125]).
• _overlay/materialize.py re-serializes the video VAE into the fused source form (bit-identical tensor values; names and fused-QKV row order differ), derives the rope.inv_freq buffer the Diffusers export drops, and patches the VAE config class names. Everything else is symlinked from the source snapshot.

Usage (no manual steps; SGLang resolves this overlay automatically):

sglang serve \
  --model-path FastVideo/FastVideo-FastH3-8-Step-V2 \
  --num-gpus 4 \
  --component-attention-backends transformer=video_sparse_attn_h3

The model inherits the MiniMax-H3 Community License from its base checkpoint.