| 文件 | 最后提交记录 | 最后更新时间 |
|---|---|---|
[pytorch][bugfix]delete configuration.json merge Co-authored-by: JZY_SC30<jiangzhiyou@huawei.com> # message auto-generated for no-merge-commit merge: !4267 merge fix_merge_problem into master [pytorch][bugfix]delete configuration.json merge Created-by: JZY_SC30 Commit-by: JZY_SC30 Merged-by: ascend-robot Description: delete merge copy of configuration.json which do not exist in hugging model file See merge request: Ascend/MindSpeed-LLM!4267 | 5 个月前 | |
[pytorch][feature]integrate moe merge process into FSDP2 training Co-authored-by: JZY_SC30<jiangzhiyou@huawei.com> # message auto-generated for no-merge-commit merge: !4250 merge fsdp2_moemerge_adapt into master [pytorch][feature]integrate moe merge process into FSDP2 training Created-by: JZY_SC30 Commit-by: JZY_SC30 Merged-by: ascend-robot Description: 当前MOE模型在FSDP2后端开启gmm特性时训练需进行一次权重转换再进行训练, 将权重转换合并入moe模型FSDP2训练流程中,实现一键拉起一体化训练。 See merge request: Ascend/MindSpeed-LLM!4250 | 5 个月前 |
| 文件 | 最后提交记录 | 最后更新时间 |
|---|---|---|
| 5 个月前 | ||
| 5 个月前 |