This website requires JavaScript.
Explore
Help
Register
Sign In
iomgaa
/
ars-opd-rebuild
Watch
1
Star
0
Fork
0
You've already forked ars-opd-rebuild
Code
Issues
Pull Requests
Actions
Packages
Projects
Releases
Wiki
Activity
Files
f1b6d1f66811a56abd130e499dc1b104da0e3729
ars-opd-rebuild
/
docs
T
History
iomgaa
2f4780b69f
docs: roadmap 存档点更新——层 2 代码关账(U1-U5 全绿),列远程冒烟三盯待办
...
Co-Authored-By: Claude Opus 4.8 (1M context) <
noreply@anthropic.com
>
2026-07-19 05:24:25 -04:00
..
00-roadmap.md
docs: roadmap 存档点更新——层 2 代码关账(U1-U5 全绿),列远程冒烟三盯待办
2026-07-19 05:24:25 -04:00
01-paper-code-map.md
docs: 修正 π̂ detach 失败机制的方向性错误(压低先验逃逸而非抬高自我强化),补充推导要点
2026-07-18 02:23:49 -04:00
02-sft-baseline.md
层1: 损失缩放终解——显式退出 HF 新式梯度累积契约(model_accepts_loss_kwargs=False)
2026-07-18 10:25:11 -04:00
03-whitebox-opd.md
层2/U1: 新增 DistillConfig(式(2) 白盒蒸馏参数)+ docs/03 表述打磨
2026-07-19 03:37:44 -04:00
appendix-claudemd-decisions.md
docs: 教学注释规范与接口回看规则入 CLAUDE.md;新增 CLAUDE.md 取舍决策附录
2026-07-17 11:59:05 -04:00
research-ideas.md
docs: 新增研究方向记录(RI-1:OmniOPD × ZO-RGD 零阶黎曼优化的三个结合点)
2026-07-18 02:57:01 -04:00