← back to paper

Review history

arxiv: 2508.20577 · 2 revisions

MERIT: Maximum-normalized Element-wise Ratio for Language Model Large-batch Training

1. 2026-08-15 CONDITIONAL MODERATE v1.4.0-alltime-deepseek-medium novelty 6.0
   152061 ms 18011 in 18559 out 2026-08-15T16:43:16.159380+00:00
2. 2026-08-05 CONDITIONAL MODERATE v1.4.0-alltime-deepseek-medium novelty 6.0
   121656 ms 18055 in 12285 out 2026-08-05T14:59:55.407646+00:00