| FazBrowse GitHub Viewer | Trending | | Home |
| Tools: [Download Repo ZIP] [Original HTTPS Page] |
Sorry, something went wrong.
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
Sorry, something went wrong.
…axolotl Run each downstream library's own bnb-specific test suite against the latest main bnb wheel (from the continuous-release_main pre-release). Catches downstream breakage before it reaches users. Jobs (all run in parallel): - test-transformers (A10G): tests/quantization/bnb/ from transformers v5.x - test-transformers-multigpu (4×L4): same, multi-GPU tests only [disabled until runner is provisioned] - test-diffusers (L40S, Docker): pytest -m bitsandbytes across quantization + model tests, using diffusers/diffusers-pytorch-cuda container - test-axolotl (A10G): kernel-level bnb tests (dequantize, LoRA, QLoRA) - test-peft (L4): PeftBnbGPUExampleTests single-GPU per Benjamin's recommendation - test-peft-multigpu (4×L4): same, multi-GPU [disabled until runner] - report (ubuntu): consolidates JUnit XMLs, posts to #bnb-daily-ci-collab via Slack bot token with threaded per-suite failure details Triggers: workflow_dispatch + pull_request (scoped to workflow/script changes). Cron schedule commented out — enable in a follow-up PR once stable. Also bumps the transformers test dep upper bound from <5 to <6, since transformers v5.x is released and the bnb test suite there has fixes (e.g. removed gated mosaicml/mpt-7b dependency). Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
| Back | FazBrowse Home | New Git URL |
Summary
Nightly integration tests that run each downstream library's own bnb-specific test suite against the latest main bnb wheel (continuous-release_main). Catches downstream breakage before it reaches users.
Replaces the legacy peft nightly-bnb.yml (deleted in peft#2858, Oct 2025) with broader coverage, now owned by the bnb repo.
Jobs
All test jobs run in parallel, followed by a consolidated report:
Runner choices match each project's own CI to avoid spurious failures from GPU-calibrated expected values.
Slack reporting
Posts a consolidated summary to #bnb-daily-ci-collab via the SLACK_CIFEEDBACK_BOT_TOKEN bot token, with threaded per-suite failure details (diffusers-style pattern). Also writes to $GITHUB_STEP_SUMMARY and uploads XMLs + markdown as artifacts.
Triggers
Other changes
Current test results
Follow-up work
🤖 Generated with Claude Code