• Joined on 2026-07-08
iomgaa pushed to main at iomgaa/ars-opd-rebuild 2026-07-18 16:51:22 +08:00
5ea58ddf59 层1/T3: data.py 数据管线——加载、messages 归一、teacher 缓存挂接、双预算 collator(docs/02 §2.2-2.3)
iomgaa pushed to main at iomgaa/ars-opd-rebuild 2026-07-18 16:40:49 +08:00
58d75cc56a 层1/T1: SFTConfig dataclass 与构造校验(docs/02 §4)
iomgaa pushed to main at iomgaa/ars-opd-rebuild 2026-07-18 16:30:57 +08:00
240f404416 工作模式变更入档:层 1 起 Claude 编写代码、用户精读提问;存档点同步
iomgaa pushed to main at iomgaa/ars-opd-rebuild 2026-07-18 16:21:32 +08:00
6c128ad6f5 docs/02:FSDP 决策定案——DDP 到 1.7B,触发点写死(4B 或 seq>8K),env 坑今日消除(脚本头部前置块)
iomgaa pushed to main at iomgaa/ars-opd-rebuild 2026-07-18 16:05:20 +08:00
0bcfc6efd4 层 0 关账:两端环境验收通过(本地 4 passed + 远程 4 passed),存档点进入层 1
iomgaa pushed to main at iomgaa/ars-opd-rebuild 2026-07-18 15:51:09 +08:00
12e2a8b0ae docs: 第二章——层1 SFT 基线开工文档(参考实现解剖+保留/替代/删除清单+任务表);CLAUDE.md 映射表补 data.py
iomgaa pushed to main at iomgaa/ars-opd-rebuild 2026-07-18 15:47:32 +08:00
77f0419bad roadmap:训练数据定为 DAPO-Math-17K(对齐论文§5.1),登记 φ/β 两处论文-代码默认值背离
iomgaa pushed to main at iomgaa/ars-opd-rebuild 2026-07-18 15:39:16 +08:00
9c0ec15716 setup_remote.sh 自清 pip 残留;存档点记录远程根分区结构性问题(/root/zym 507G)
iomgaa pushed to main at iomgaa/ars-opd-rebuild 2026-07-18 15:30:37 +08:00
991e277316 CLAUDE.md:日志不缓存升级为全局硬规则(含 conda run 事故伤疤标注)
iomgaa pushed to main at iomgaa/ars-opd-rebuild 2026-07-18 15:28:16 +08:00
b679bcb24f 修复 setup_remote.sh:弃用 conda run(输出整体缓冲),改为直调环境内 pip/python 实现实时日志
iomgaa pushed to main at iomgaa/ars-opd-rebuild 2026-07-18 15:21:03 +08:00
f255d5613a docs: roadmap 增加存档点小节(断点恢复用),登记层 0 当前状态
iomgaa pushed to main at iomgaa/ars-opd-rebuild 2026-07-18 15:15:25 +08:00
270a9ab291 层0:统一依赖清单 + 远程一键搭建脚本 + 密钥模板
cccccc0ce6 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)
72799a34dc tests: 预置 π̂ detach 命门守护测试(对应 docs/01 §3.4,层 3 接入真实 estimator)
c9eddd5be8 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
d0bedd564c 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
Compare 5 commits »
iomgaa pushed to main at iomgaa/ars-opd-rebuild 2026-07-17 23:54:12 +08:00
8647c5a89d docs: 第一章——论文精读与参考实现全景映射(含实现与论文差异清单)
6dc13314d8 仓库骨架:README、CLAUDE.md、分层重构路线图(docs/00-roadmap.md)
iomgaa created branch main in iomgaa/ars-opd-rebuild 2026-07-17 23:54:12 +08:00
iomgaa created repository iomgaa/ars-opd-rebuild 2026-07-17 23:17:09 +08:00
iomgaa pushed to main at iomgaa/Video-Tree-TRM5 2026-07-17 18:13:53 +08:00
6ac96dff64 Merge branch 'feat/question-gen-v3'
3dcc7d9899 fix: use SLURM_SUBMIT_DIR to locate repo root under sbatch
Compare 2 commits »
iomgaa pushed to feat/question-gen-v3 at iomgaa/Video-Tree-TRM5 2026-07-17 18:13:44 +08:00
3dcc7d9899 fix: use SLURM_SUBMIT_DIR to locate repo root under sbatch
iomgaa pushed to feat/question-gen-v3 at iomgaa/Video-Tree-TRM5 2026-07-17 18:06:49 +08:00
9bb09a44d4 feat: add Slurm gpu-sbatch training entry for 4090-server
iomgaa pushed to main at iomgaa/Video-Tree-TRM5 2026-07-17 18:06:32 +08:00
7386e5af57 Merge remote-tracking branch 'origin/main'
0414948682 merge: question-gen v3 + preflight fixes + continuous concurrent gate
9bb09a44d4 feat: add Slurm gpu-sbatch training entry for 4090-server
b0be1f1ae5 docs: gate speedup final review and delivery record
172b7a8ec4 merge: continuous concurrent gate speedup (algo #6 semantic revision, design v3)
Compare 400 commits »
iomgaa created branch feat/question-gen-v3 in iomgaa/Video-Tree-TRM5 2026-07-17 17:43:52 +08:00