nodes/hao-ai-lab-FastVideo
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[misc] use the full release name FastVideo-Minimax-FastH3-Preview-v0.1

Browse Source 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
Will Lin
2026-08-21 09:27:38 +00:00
co-authored by Claude Fable 5
parent 37aa382cce
commit 9713ea1275
2 changed files with 4 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files
examples/inference/basic/README.md
+1 -1
View File
@@ -37,7 +37,7 @@ For the few-step (4-step, DMD2-distilled) MiniMax-H3 preview, generating synchro
```
python examples/inference/basic/basic_fasth3.py --prompt "your prompt" [--vsa-sparsity 0.9]
```
The default checkpoint `FastVideo/FastH3-Preview-v0.1` is private on the Hub while its license review completes; until it flips public, pass `--model-path` with a local snapshot of the release.
The default checkpoint `FastVideo/FastVideo-Minimax-FastH3-Preview-v0.1` is private on the Hub while its license review completes; until it flips public, pass `--model-path` with a local snapshot of the release.
On Blackwell (sm_100) GPUs with a `fastvideo-kernel` build that carries the sm_100a block-sparse extension, `--vsa-kernel sm100a` routes the tile-64 attention forwards through the CUDA kernel instead of Triton (it sets `FASTVIDEO_VSA_SM100A=1` before the pipeline boots); if the extension or the arch is missing, the run warns once and falls back to Triton.
examples/inference/basic/basic_fasth3.py
+3 -3
View File
@@ -1,7 +1,7 @@
# SPDX-License-Identifier: Apache-2.0
"""Few-step video+audio generation with the DMD2-distilled MiniMax H3 preview.
FastVideo/FastH3-Preview-v0.1 is a 4-step distillation of
FastVideo/FastVideo-Minimax-FastH3-Preview-v0.1 is a 4-step distillation of
MiniMaxAI/MiniMax-H3 (data-free DMD2): it walks a 4-step grid on the release
sampler's shift-12 schedule instead of the base model's 50 steps, generating
synchronized video and audio in one pipeline call.
@@ -35,11 +35,11 @@ from fastvideo.api import (
def parse_args() -> argparse.Namespace:
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("--model-path", default="FastVideo/FastH3-Preview-v0.1")
parser.add_argument("--model-path", default="FastVideo/FastVideo-Minimax-FastH3-Preview-v0.1")
# The HF repo is private while the MiniMax H3 Community License review
# completes; until it flips public, pass --model-path with a local
# snapshot of the release instead (e.g. the team export at
# /mnt/lustre/vlm-wlsaidhi/fastvideo/exports/FastH3-Preview-v0.1).
# /mnt/lustre/vlm-wlsaidhi/fastvideo/exports/FastVideo-Minimax-FastH3-Preview-v0.1).
parser.add_argument("--prompt", required=True)
parser.add_argument("--output", default="outputs/fasth3")
parser.add_argument("--height", type=int, default=768)