← back to paper

Review history

arxiv: 2608.03467 · 2 revisions

When Correct Solutions Repeat: Rarity-Aware Credit Redistribution for GRPO

1. 2026-08-08 CONDITIONAL MODERATE v1.4.0-alltime-deepseek-medium novelty 5.0
   167523 ms 19024 in 17321 out 2026-08-08T00:45:46.764278+00:00
2. 2026-08-05 CONDITIONAL MODERATE v1.4.0-daily-deepseek-medium novelty 4.0
   181016 ms 11700 in 16288 out 2026-08-05T18:37:38.354277+00:00