diff options
| author | Yuren Hao <yurenh2@illinois.edu> | 2026-08-05 06:11:06 -0500 |
|---|---|---|
| committer | Yuren Hao <yurenh2@illinois.edu> | 2026-08-05 06:11:06 -0500 |
| commit | 044973e2a78e0faa1296ac370cefbfc2f6feadd6 (patch) | |
| tree | fb8fd3df6eaffb36e20757c5f28bce6046b8c607 /ep_run/FUGU_VERDICT_FULL.md | |
| parent | 98e145cda386167ce6d9d20120b4e265b3a6471a (diff) | |
优化器价目表实装: Lion/OLion/Adafactor2D/build_alt + --opt 六选项
OLion按论文算法框(慢动量+Nesterov混合→NS→逐元素sign→γ=0.2定标+解耦wd; betas取Lion默认
0.9/0.99并注明为假设); Adafactor2D=行/列因子化二阶矩+RMS-1裁剪(07-11审计的"模拟原生Adam");
build_alt与build_hybrid同分割(矩阵走被测规则+其余AdamW), sgdm例外全参数SGD(全局部基线)。
四路CPU冒烟全过(损失下降, gate cos 1.0000)。
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
Diffstat (limited to 'ep_run/FUGU_VERDICT_FULL.md')
0 files changed, 0 insertions, 0 deletions