docs: add Sage Attention speedup section to MiniMax H3 tutorial - #1348
Conversation
📝 WalkthroughWalkthroughChangesMiniMax H3 Sage Attention
Possibly related PRs
🚥 Pre-merge checks | ✅ 2✅ Passed checks (2 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
✨ Simplify code
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
There was a problem hiding this comment.
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
tutorials/video/minimax/minimax-h3.mdx (1)
50-50: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick winScope the Sage Attention speed claim to the reported benchmark.
The PR objective identifies the reported generation improvement at 720p. The current text presents the roughly twofold speedup without that condition. Scope every sentence to 720p and regenerate translation metadata after the text changes.
- tutorials/video/minimax/minimax-h3.mdx#L50-L50: scope the English claim to 720p.
- ja/tutorials/video/minimax/minimax-h3.mdx#L65-L65: scope the Japanese claim to 720p and refresh its translation metadata.
- ko/tutorials/video/minimax/minimax-h3.mdx#L66-L66: scope the Korean claim to 720p and refresh its translation metadata.
- zh/tutorials/video/minimax/minimax-h3.mdx#L66-L66: scope the Chinese claim to 720p and refresh its translation metadata.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@tutorials/video/minimax/minimax-h3.mdx` at line 50, Scope the Sage Attention speedup statement to 720p in tutorials/video/minimax/minimax-h3.mdx:50-50, ja/tutorials/video/minimax/minimax-h3.mdx:65-65, ko/tutorials/video/minimax/minimax-h3.mdx:66-66, and zh/tutorials/video/minimax/minimax-h3.mdx:66-66; after updating the translated claims, regenerate the translation metadata for the Japanese, Korean, and Chinese files.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Outside diff comments:
In `@tutorials/video/minimax/minimax-h3.mdx`:
- Line 50: Scope the Sage Attention speedup statement to 720p in
tutorials/video/minimax/minimax-h3.mdx:50-50,
ja/tutorials/video/minimax/minimax-h3.mdx:65-65,
ko/tutorials/video/minimax/minimax-h3.mdx:66-66, and
zh/tutorials/video/minimax/minimax-h3.mdx:66-66; after updating the translated
claims, regenerate the translation metadata for the Japanese, Korean, and
Chinese files.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: ASSERTIVE
Plan: Pro Plus
Run ID: a336a596-c03a-4e4f-a0c9-647d2857d70e
📒 Files selected for processing (4)
ja/tutorials/video/minimax/minimax-h3.mdxko/tutorials/video/minimax/minimax-h3.mdxtutorials/video/minimax/minimax-h3.mdxzh/tutorials/video/minimax/minimax-h3.mdx
Add a new
Speeding up generation with Sage Attentionsection to the open-weights MiniMax H3 tutorial page.Sage Attention roughly doubles generation speed at 720p with minimal quality loss. The section covers:
sageattentionpackage themselves (wheel from the SageAttention releases page matching their PyTorch/CUDA) plus the KJNodes custom nodes (Patch Sage Attention KJ)Patch Sage Attention KJbetweenUNETLoaderandBasicGuider(model in / model out), setsage_attentiontoauto; only the guider needs the patch, not the scheduler--use-sage-attnlaunch flagAll 4 language files updated with manual sync + hash sync.