srm.git/runs, branch main Unnamed repository; edit this file 'description' to name the repository. Add HRM-Orth v1 (codex round 2 Q6 pivot) 2026-05-23T17:04:04+00:00 YurenHao0426 Blackhao0426@gmail.com 2026-05-23T17:04:04+00:00 fe4d92760f9d9ce9d9f41eb0fe69dd9eadc1534c Patch HRM Block with Lipschitz-bounded ops: - attention → cosine-normalized softmax attn - SwiGLU → OrthLinear (Cayley + weak diag scale) + MaxMin + OrthLinear - rms_norm + add → weighted residual (1-σ(w))·h + σ(w)·f(h) - Weak orthogonality: diag(s) with s_i ∈ [0.95, 1.0] for compression directions Keeps HRM ACT framework + H_level/L_level + cycles unchanged. Predicted +5-7pp vs SRM v1 (codex Q5 decomp): +1.5-2.5 (remove ReLU rank-kill via MaxMin) +2.0-3.0 (remove AOL attenuation via Cayley) +1.0-1.5 (orthogonal residual flow) Also adds: train_hrm_orth.py trainer, SRM v1 run logs, .gitignore ckpts/.codex Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> 
Patch HRM Block with Lipschitz-bounded ops:
- attention → cosine-normalized softmax attn
- SwiGLU → OrthLinear (Cayley + weak diag scale) + MaxMin + OrthLinear
- rms_norm + add → weighted residual (1-σ(w))·h + σ(w)·f(h)
- Weak orthogonality: diag(s) with s_i ∈ [0.95, 1.0] for compression directions

Keeps HRM ACT framework + H_level/L_level + cycles unchanged.

Predicted +5-7pp vs SRM v1 (codex Q5 decomp):
  +1.5-2.5 (remove ReLU rank-kill via MaxMin)
  +2.0-3.0 (remove AOL attenuation via Cayley)
  +1.0-1.5 (orthogonal residual flow)

Also adds: train_hrm_orth.py trainer, SRM v1 run logs, .gitignore ckpts/.codex

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>