Evaluation workbench
Evaluate RMSNorm correctness.
Run Training Day's RMSNorm CPU check against a reference implementation, then inspect the result.
Live evaluation
RMSNorm correctness
Run is enabled only after the service confirms the fixed CPU drill and limits.
Ready
Waiting for a run
Connect to the service, then run the fixed correctness evaluation.
- Correctness
- Cases evaluated
- Evaluation wall time
- Redline policy
- GPU performance
- GPU time / API cost
- Run ID
- Completed at
Available after a completed evaluation.
Returned by the evaluator
Provenance
SHA-256
- Candidate
- Reference
- Tests
- Drill
- Policy
- Case manifest