ci: enforce pre-commit style check via GitHub Actions

[Anerudhan]

What

Adds .github/workflows/pre-commit.yml: a Style check workflow that runs pre-commit run --all-files on every PR and on pushes to develop/main, on a plain ubuntu-latest runner (no GPU). Hook environments are cached keyed on the config hash, so steady-state runs take seconds.

This enforces in CI exactly what CONTRIBUTING.md already asks contributors to run locally — clang-format v21.1.6 for C++/CUDA and black 26.3.1 (line length 160) for Python/notebooks, as pinned in .pre-commit-config.yaml — replacing the internal analysis:clang-format GitLab job (removed separately), which used unpinned tool versions from the CI image and occupied a GPU runner for a pure text check.

Included cleanup

• One-time black fix for the 3 files the hook currently flags (benchmark/attention_inference/benchmark_single_attention.py, benchmark/attention_inference/configs/qwen3vl_vit.py, test/python/sdpa/frost/test_sdpa_fwd_dsl_sm100.py) so the check passes from day one. benchmark/ was never covered by the old script, which is how these drifted.
• Fix a stale frost README reference to the internal-only ci/run_style_check_diff.sh (points to pre-commit run --all-files now).

Verification

pre-commit run --all-files passes locally on this branch (clang-format, black, black-jupyter all green). The workflow itself will run on this PR.

🤖 Generated with Claude Code

Summary by CodeRabbit

• Chores

  • Added automated formatting checks for pull requests and pushes to the primary development branches.
  • Formatting checks now provide clear, colorized differences when validation fails.

• Documentation

  • Updated contributor guidance to use the current pre-commit formatting workflow.

• Style

  • Standardized formatting across attention benchmarks, vision configuration files, and SDPA tests without changing runtime behavior.

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 5a9ff017-5536-45de-ab53-c424f896a92e

📥 Commits

Reviewing files that changed from the base of the PR and between 36a06c4 and 6184c1c.

📒 Files selected for processing (5)

• .github/workflows/pre-commit.yml
• benchmark/attention_inference/benchmark_single_attention.py
• benchmark/attention_inference/configs/qwen3vl_vit.py
• python/cudnn/frost/README.md
• test/python/sdpa/frost/test_sdpa_fwd_dsl_sm100.py

---

📝 Walkthrough

Walkthrough

The change adds a pre-commit GitHub Actions workflow, updates the FROST formatting command documentation, and applies formatting-only changes to benchmark, configuration, and test files.

Changes

Pre-commit alignment

Layer / File(s)	Summary
Pre-commit CI and contributor workflow .github/workflows/pre-commit.yml, python/cudnn/frost/README.md	The workflow runs all pre-commit hooks on pull requests and pushes to develop or main. The README documents the same command.
Formatting normalization benchmark/attention_inference/..., test/python/sdpa/frost/test_sdpa_fwd_dsl_sm100.py	Formatting changes preserve benchmark arguments, configuration values, test arguments, and output behavior.

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

Suggested reviewers: vedaanta

✨ Finishing Touches 💡 1

🛠️ Fix failing CI checks 💡

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

_{Comment @coderabbitai help to get the list of available commands.}