[CI report] Upload the vLLM / SGLang CI reports to the Hub and post a summary to Slack (#50896)
* [CI report] Upload the vLLM / SGLang CI reports to the Hub and post a summary to Slack
- analyze_ci_report.py: also write job_results.json (run metadata + per-job status, test
counts and failures, like model_results.json of the daily CI). It also lists failed jobs
without a pytest summary (killed, setup failure, log unavailable) as "no_summary".
- analyze_gh_wf_runs.py: extract_test_counts() parses pytest's final line(s).
- New utils/ci_report_notify.py: uploads the reports in one commit to a Hub dataset repo
(<date>/ for scheduled runs, <date>/runs/<run_number>-<run_id>/ otherwise, like the daily
CI) and posts a summary to Slack.
- vLLM / SGLang workflows: new step in the report job, to hf-internal-testing/transformers_{vllm,sglang}_ci
and #transformers-ci-{vllm,sglang}.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* [CI report] Slack: drop the Previous run button, show New failures before All results
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* [CI report] Slack: escape text from the logs (&, <, >) in the message
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* [vLLM CI] Only give HF_TOKEN to the jobs that run the tests
Move it from the workflow-level env to the 4 CPU test jobs (the GPU job already sets its
own), so the report job no longer gets it. Same as the SGLang CI (#50878).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>