nano-gpt-data-curation
gpt-5.6-sol
nano-gpt-data-curation — graded by the task's verifier against a measured baseline.
- backend
- modal
- variant
- codex azure_ai/gpt-5.6-sol
- metric
- held-out multi-domain perplexity
- actions
- 155
- duration
- 45m
- recorded
- 2026-08-03
run iddclm-nano-curation__gfZ879v
Score
reward
398.670
Summary
Completed all deliverables:
- selection.json — 20,868 unique valid IDs; verified 12M-token coverage.
- curate.py — reproducibly regenerates the exact selection.
- claim.md — hypothesis, mechanism, falsification, and transfer.
Best dev perplexity: 420.10, versus 485.76 for random selection.