Skip to content

Conversation

@Nottlespike
Copy link
Contributor

Summary

Syncs all changes from the AlphaHENG/contrib/metal_marlin folder to this repository.

Major Changes

  • Benchmarks: New profiling and throughput testing scripts
  • Metal Shaders: Gated MLP fusion implementation
  • Tests: Fast-path integration tests
  • Utilities: Library loading test utilities
  • Core Improvements:
    • Buffer pool enhancements
    • Encoder cache updates
    • MoE manager improvements
  • Trellis Module:
    • Attention layer enhancements
    • Dispatch improvements
    • Linear layer optimizations
  • Various kernel optimizations and bug fixes across Metal shaders

Major changes:
- Benchmark scripts for profiling and throughput testing
- Gated MLP Metal shader fusion implementation
- Fast-path integration tests
- Library loading test utilities
- Buffer pool, encoder cache, MoE manager improvements
- Trellis module enhancements (attention, dispatch, linear layers)
- Various kernel optimizations and bug fixes
@Nottlespike Nottlespike merged commit 2c9c0e1 into main Feb 3, 2026
2 of 4 checks passed
@Nottlespike Nottlespike deleted the updates-from-alphaheng branch February 3, 2026 20:18
Copy link

@chatgpt-codex-connector chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: a6877bd7c4

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

Comment on lines +5 to +9
paths:
- 'src/*.metal'
pull_request:
paths:
- 'src/*.metal'

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2 Badge Expand workflow paths beyond src/*.metal

This workflow only triggers on src/*.metal, so shader changes in nested directories (e.g., src/fusion/*.metal, src/vision/*.metal, etc.) won't run the compile/audit checks at all. That means newly added or modified shaders outside the root src folder can land without being compiled or audited, letting breakage slip into main. Consider expanding the path filter (and the compile loop) to cover all Metal shaders under src/**.metal.

Useful? React with 👍 / 👎.

Comment on lines +30 to +32
- name: Run Kernel Audit
run: |
output=$(uv run python audit_kernels.py)

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P1 Badge Fix audit_kernels.py path in workflow

uv run python audit_kernels.py runs from the repo root, but the only audit_kernels.py in this tree lives under developer_tests/ (verified with rg --files -g 'audit_kernels.py'). As written, the step will fail with “No such file or directory” whenever the workflow runs. Point the command at the correct path (e.g., developer_tests/audit_kernels.py) or cd before invoking it.

Useful? React with 👍 / 👎.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant