[leejet]

Summary

• Split MiniMax-H3 MLP computation along the token axis to avoid oversized F32 activations triggering Vulkan CPU fallback.
• Account for batch dimensions and buffer alignment, preserving the existing path for inputs that fit.

Related Issue / Discussion

Addresses #2242.

Additional Information

N/A

Checklist

• I have read and confirmed this PR follows the contribution guidelines.

Summary by CodeRabbit

• Improvements
  • Large model inputs are now processed in smaller portions when needed to fit backend buffer limits, while inputs that fit continue to be processed in one pass. This helps the model handle constrained backend environments more reliably.

[coderabbitai]

No actionable comments were generated in the recent review.

Recent review info

Run configuration

• Source mode: Auto
• Repository: Unavailable
• Branch: Unavailable
• File: No configuration file from this repository applied
• Commit: Unavailable
• Layers: Repository UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: 3d5e0262-0b92-415e-b0e2-6c78448665e3

Commits

Reviewing files that changed from the base of the PR and between 45a63ac and 792cdfa.

Files selected for processing: 1

• src/model/diffusion/minimax_h3.hpp

Included review availability: This review used your included allowance. Your plan provides up to 8 included reviews per hour; 6 remain after this review.

Walkthrough

The MLP stores its feed-forward hidden size and uses the backend maximum buffer size to determine token-axis chunking. Oversized inputs are processed in contiguous chunks and concatenated. Inputs within the limit, or inputs without a backend, retain single-pass processing.

Changes: MiniMax-H3 MLP backend-aware chunking

Layer / File(s)	Summary
Backend-aware MLP forward pass src/model/diffusion/minimax_h3.hpp	MLP stores ffn_hidden_size. When a backend is present, forward derives a chunk limit from the aligned maximum buffer size and activation widths. Oversized inputs are sliced along dimension 1, made contiguous when needed, processed, and concatenated. Other inputs use one pass.

Estimated code review effort: 2 (Simple) | ~10 minutes

Change: Bug fix

Possibly related issues

• [Bug] MiniMax-H3 DiT: MLP fc1 output exceeds Vulkan's 4 GiB buffer limit above 37,449 tokens; the MLP of every layer falls back to CPU #2242: This change chunks MiniMax-H3 MLP activations along the token axis to keep them within the backend buffer limit described in the issue.

Possibly related PRs

• leejet/stable-diffusion.cpp#2240: This PR extends the MiniMax-H3 MLP implementation changed in PR #2240, adding token-axis chunking when backend buffer limits would be exceeded.

Merge Risk: Minimal · up to 792cd

No concrete merge-blocking issue was established for the chunked MLP.

Pre-merge checks: 4 · 1

Failed checks: 1 warning

Check name	Status	Explanation	Resolution
Docstring Coverage		Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff with cyclomatic complexity of at least 2. Documentat…	Write docstrings for the functions missing them to satisfy the coverage threshold.

Passed checks: 4 passed

Check name	Status	Explanation
Title check		The title clearly and concisely describes the main change: chunking MiniMax-H3 MLP computation to respect backend buffer limits.
Description check		The description includes all required sections. It explains the change and rationale, references issue #2242, provides additional information, and confirms the contribution checklist.
Linked Issues check		Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check		Check skipped because no linked issues were found for this pull request.

Full details: Docstring Coverage

Explanation

Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff with cyclomatic complexity of at least 2. Documentation-only edits and deletions within surviving functions are included; unchanged diff context and fully deleted functions are excluded. Analyzed 1 functions across 1 files.

Resolution

Write docstrings for the functions missing them to satisfy the coverage threshold.

Finishing Touches

Generate docstrings

• Commit to this branch
• Create a new PR

Generate unit tests (beta)

• Commit to this branch
• Create a new PR

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

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.

Share

• X
• Mastodon
• Reddit
• LinkedIn