Pith. sign in

← back to paper

Review history

arxiv: 2608.09417 · 2 revisions

Why Post-Norm Transformers Collapse: Attention Amplification and Gradient Repair Failure

  1. 2026-08-14 CONDITIONAL HIGH v1.4.0-alltime-deepseek-medium novelty 6.0
    206273 ms 34543 in 23995 out 2026-08-14T04:15:35.115498+00:00
  2. 2026-08-11 CONDITIONAL MODERATE v1.4.0-daily-deepseek-medium novelty 6.0
    122134 ms 34543 in 12033 out 2026-08-11T17:43:24.604754+00:00