Skip to content

(docs): enchance roll_docs & update local code test. - #142

Merged
PanAndy merged 7 commits into
mainfrom
sync/sync_to_github_0817
Aug 17, 2025
Merged

(docs): enchance roll_docs & update local code test.#142
PanAndy merged 7 commits into
mainfrom
sync/sync_to_github_0817

Conversation

@PanAndy

@PanAndy PanAndy commented Aug 17, 2025

Copy link
Copy Markdown
Collaborator

(docs): enchance roll_docs.
(feat): update local code test.
(feat): add qwen3 8B&235B config.

chocoded and others added 7 commits August 17, 2025 22:39
修改了保存检查点的条件,使其在达到最大步数时也进行保存。
Link: https://code.alibaba-inc.com/openlm/ScaleAligner/codereview/22963401
* refactor: update checkpoint condition logic in base pipeline

* refactor: adjust checkpoint condition logic

(cherry picked from commit 34ab810bf9e9a65ecd2d38b1a602872902918b8a)
在 `frozen_lake` 和 `sokoban` 环境的配置文件中,启用了 KL 散度损失并调整了优势标准化和裁剪参数,同时修正了格式惩罚值。
Link: https://code.alibaba-inc.com/openlm/ScaleAligner/codereview/22978205
* (fix): fix gigpo config.

(cherry picked from commit ced1143026ffa5099a8b5d889e578555e48e0b57)
Discussed-in: Merge-Request 22959728 1/2, URL: https://code.alibaba-inc.com/openlm/ScaleAligner/codereview/22959728
(cherry picked from commit cdb33c956a86f9093f1ba8f79945b040182661bf)
Discussed-in: Merge-Request 22959728 2/2, URL: https://code.alibaba-inc.com/openlm/ScaleAligner/codereview/22959728
(cherry picked from commit 60f07be534d5e215e098db8b5e7d058e15efce5e)
文档更新将 Megatron 模型的相关说明更改为 MCoreAdapter 模型,并明确指出转换工具适用于由 `megatron_strategy` 训练的模型。
Link: https://code.alibaba-inc.com/openlm/ScaleAligner/codereview/23069812
* (docs): fix mcore adapter convert.

(cherry picked from commit 44d354363baafc9c8637f6c963696196b1dba95f)
@PanAndy
PanAndy requested a review from kkkky123 August 17, 2025 15:21
@PanAndy
PanAndy merged commit 1f66a59 into main Aug 17, 2025
6 checks passed
@PanAndy
PanAndy deleted the sync/sync_to_github_0817 branch August 17, 2025 15:25
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants