| 文件 | 最后提交记录 | 最后更新时间 |
|---|---|---|
新增支持模型Deepseekv3.2/exp/speciale Co-authored-by: Minds66<dengshumin2@huawei.com> # message auto-generated for no-merge-commit merge: !915 merge dpskv32 into master 新增支持模型Deepseekv3.2/exp/speciale Created-by: Minds66 Commit-by: Minds66 Merged-by: ascend-robot Description: What this PR does / why we need it? We support new model Deepseekv3.2/exp/speciale Does this PR introduce any user-facing change? Update deepseev32 readme doc in verl_npu/docs How was this patch tested? Please explain how to verify the correctness and effectiveness of this feature, as well as its usage constraints and limitations. See merge request: Ascend/MindSpeed-RL!915 | 4 个月前 | |
新增支持模型Deepseekv3-671b Co-authored-by: Minds66<dengshumin2@huawei.com> # message auto-generated for no-merge-commit merge: !887 merge dpsk671b-v3.1 into master 新增支持模型Deepseek671b-v3 Created-by: Minds66 Commit-by: Minds66 Merged-by: ascend-robot Description: 新增NPU支持模型Deepseekv3-671b，新增mbridge支持deepseek671b在线权重转换，gsm8k数据集上分数稳定上升  See merge request: Ascend/MindSpeed-RL!887 | 6 个月前 | |
添加verl-sglang脚本和指导说明 Co-authored-by: xiazhahe<nieshiyu1@huawei.com> # message auto-generated for no-merge-commit merge: !877 merge master into master 添加verl-sglang脚本和指导说明 Created-by: xiazhahe Commit-by: xiazhahe Merged-by: ascend-robot Description: 添加verl-sglang脚本和指导说明 See merge request: Ascend/MindSpeed-RL!877 | 6 个月前 | |
修改gptoss、int8 rollout文档 Co-authored-by: nuerxiati<738457498@qq.com> # message auto-generated for no-merge-commit merge: !891 merge update_int8_gptoss into master 修改gptoss、int8 rollout文档 Created-by: NurxatAbilmit Commit-by: nuerxiati Merged-by: ascend-robot Description: update int8 gptoss See merge request: Ascend/MindSpeed-RL!891 | 6 个月前 | |
[fix]: fix dpskv32 patches and docs Co-authored-by: Minds66<dengshumin2@huawei.com> # message auto-generated for no-merge-commit merge: !922 merge dpskv32-2 into master [fix]: fix dpskv32 patches and docs Created-by: Minds66 Commit-by: Minds66 Merged-by: ascend-robot Description: What this PR does / why we need it? Fixed the issue where training failed when directly following the instructions in deepseekv32.md. Does this PR introduce any user-facing change? Nope How was this patch tested? Please explain how to verify the correctness and effectiveness of this feature, as well as its usage constraints and limitations. See merge request: Ascend/MindSpeed-RL!922 | 4 个月前 | |
新增NPU GSPO脚本 master Co-authored-by: Hcm03<chengminhua1@huawei.com> # message auto-generated for no-merge-commit merge: !778 merge master into master 新增NPU GSPO脚本 master Created-by: Hcm03 Commit-by: Hcm03 Merged-by: ascend-robot Description: 新增NPU GSPO脚本 master 具体可见：https://wiki.huawei.com/domains/140211/wiki/276605/WIKI202511279180497 See merge request: Ascend/MindSpeed-RL!778 | 7 个月前 | |
support retool on npu Co-authored-by: yaochao20<yaochao20@huawei.com> # message auto-generated for no-merge-commit merge: !874 merge pr-only-yaochao20-v3 into master support retool on npu Created-by: yaochao20 Commit-by: yaochao20 Merged-by: ascend-robot Description: support retool on npu See merge request: Ascend/MindSpeed-RL!874 | 6 个月前 |