/eval · /metrics
Eval hub
Held-out compare first (classic · flat · modern · BASE), then Prior SOAP six-metric quality board. Auto-refresh; never invents LoRA scores; FT create stays paused. Metric guide on /eval-methods; live side-by-side on /lab.
Last updated: —
Poll: ~5s · loading…· ok
Loading eval results…
Held-out compare · classic · flat · modern · BASE
Missing reports/EVAL_HELDOUT_THREE_ARM.json. Run npm run eval:heldout:three-arm.
Prior SOAP six-metric board
Missing reports/EVAL_HELDOUT_SIX_METRIC.json. Run npm run eval:heldout:six-metric.
Routes: /eval · /metrics · /eval-methods · API GET /api/metrics · sources EVAL_HELDOUT_THREE_ARM.json · EVAL_HELDOUT_SIX_METRIC.json