Model Trace Viewer (Base vs SFT)
Problem ID
Base Model
—
SFT Model
—
Metrics (averaged over all rollouts)
Computed from text lengths; response_length used when present.
Max children per fork
Averaged across all rollouts
Parent + longest descendant chain (avg chars)
(I.e. the critical path length)
Context lengths
Base JSONL file
SFT JSONL file
Optional: Metrics CSV (any task)
Loads into metrics panel; also logs metrics to console.