msverl-daily-regression-triage

To automate regression triage for daily msverl runs, provide the path to the latest comparison log. The skill parses logs and results to identify pass, train_error, or accuracy_regression status, then ranks suspect commits with evidence-based scoring.

Commit ranking for training failure evidence extraction works by collecting recent commits from verl main and MindSpeed master within a defined time window, then scoring suspects using evidence-driven logic to pinpoint the most likely culprit commits.

The best way to identify accuracy regression in verl main and MindSpeed master is by parsing baseline comparison logs. The process extracts relevant failure evidence and produces a ranked list of candidate commits with recommended next steps.

Yes, you can use log analysis to find a suggested fix direction for training errors. The triage workflow extracts the failure tail from daily training logs and generates concise remediation guidance alongside the ranked suspect commits.

The report after triaging a daily msverl regression run includes the health status, time window, extracted evidence, candidate commits, confidence levels, and a suggested fix direction to guide remediation.

Automated regression triage does not require manual git commit inspection. The skill automatically collects recent commits from verl main and MindSpeed master, ranks them with evidence-based scoring, and outputs a concise triage report.