Pull Requests · Ascend/MindSpeed-RL - Gitee.com

开源项目 > 人工智能 > 机器学习/深度学习 &&

全共卡模式支持断点续训单独加载ref权重可合并

ci-pipeline-passed

ascend-cla/yes

liuxinyang:dev Ascend:dev

!259 11

2025-04-28 19:45

修复RL仓reuse_fp32_param特性不能用的问题可合并

ascend-cla/yes

liuxinyang:lyf Ascend:dev

!258 4

2025-04-28 19:06

特性use_remove_padding

ascend-cla/yes

草稿

Nurxat:dev Ascend:dev

!255 1

2025-04-27 22:32

dataset shuffle逻辑修复与变更可合并

ci-pipeline-passed

ascend-cla/yes

stat/needs-squash

LookAround:fix_dataset Ascend:dev

!251 102

2025-04-27 15:00

增加CoC功能可合并

ascend-cla/yes

yuqi:coc Ascend:dev

!249 1

2025-04-25 16:26

dev add profiler 可合并

ci-pipeline-passed

ascend-cla/yes

lgtm

tangmengcheng:feature_dev_add_profiler Ascend:dev

!240 41

2025-04-22 15:20

新增partial rollout特性存在冲突

ci-pipeline-failed

ascend-cla/yes

stat/needs-squash

LookAround:partial_rollout Ascend:dev

!237 11

2025-04-21 18:14

reward 存在冲突

ascend-cla/yes

朱意霖:dev Ascend:dev

!234 1

2025-04-21 15:31

特性：use_remove_padding

ascend-cla/yes

草稿

Nurxat:master Ascend:master

!233 13

2025-04-21 09:27

feat: 支持td profiling 可合并

ci-pipeline-failed

ascend-cla/yes

stat/needs-squash

sunjianping:dev Ascend:dev

!231 61

2025-04-20 12:04

1

https://gitee.com/ascend/MindSpeed-RL.git

git@gitee.com:ascend/MindSpeed-RL.git

ascend

MindSpeed-RL

MindSpeed-RL