GPU concurrency #399
GPU concurrency #399
Conversation
Coverage report
This PR does not seem to contain any modification to coverable code. |
There was a problem hiding this comment.
Pull request overview
This PR implements dynamic GPU runner selection to enable concurrent GitHub Actions workflow runs across multiple GPUs. Instead of using a single fixed runner, the workflow now distributes jobs across 5 different GPU runners using a round-robin selection based on the GitHub run number.
Changes:
- Added a
select-runnerjob that picks one of 5 available GPU runners using modulo-based distribution - Updated the main
runjob to use the dynamically selected runner and added concurrency controls per runner
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| runner: ${{ steps.pick.outputs.runner }} | ||
| steps: | ||
| - id: pick | ||
| run: | |
There was a problem hiding this comment.
The runner list skips gpu1, gpu3, and gpu6 without explanation. Consider adding a comment explaining why these specific GPUs are selected, or use a more consistent naming pattern if possible.
| run: | | |
| run: | | |
| # Only schedule jobs on GPUs that are available for CI use. | |
| # gpu1, gpu3, and gpu6 on b200-02 are reserved/unavailable, so they are intentionally excluded. |
| - id: pick | ||
| run: | | ||
| runners=("b200-02-gpu0" "b200-02-gpu2" "b200-02-gpu4" "b200-02-gpu5" "b200-02-gpu7") | ||
| index=$(( ${{ github.run_number }} % 5 )) |
There was a problem hiding this comment.
The modulo value is hardcoded as 5. If runners are added or removed from the array, this must be manually updated. Consider using ${#runners[@]} to automatically match the array length.
| index=$(( ${{ github.run_number }} % 5 )) | |
| index=$(( ${{ github.run_number }} % ${#runners[@]} )) |
No description provided.