Merge blocked
【openlibing.ci】识别到代码检查告警抑制注释，匹配工具：pylint，请Committer检视其合理性。
【openlibing.ci】识别到代码检查告警抑制注释，匹配工具：pylint，请Committer检视其合理性。
【openlibing.ci】识别到代码检查告警抑制注释，匹配工具：pylint，请Committer检视其合理性。
Thanks for your pull-request.
The full list of commands accepted by me can be found at here.
You can get sig-info at here.
You can self-configure the PR merge rules for this repository. For more details, please refer to Here.
PR Approval Progress
⚠️ This PR does not yet meet the following requirements:lgtm (requires ≥ 2 person(s) per module)、approve (requires ≥ 1 person(s) per module)
Module Approval Details
| module | lgtm status | approve status |
|---|---|---|
| repo-Ascend/MindSpeed-Ops | ❌ (0/2)(You can also ask: guihaowen666, tichang, 周蓓蓉, 雷珍珍, HANHU1CHEN) | ❌ (0/1)(You can also ask: 刘荣, 华郁秀, 周蓓蓉, 雷珍珍, 郑加利) |
💡 Tip:
- Committer can comment
/approveor/lgtm- Commenting
/approveimplies both code review (lgtm) and intent to merge (approve)
CLA Signature Pass
xiaomaomao666, thanks for your pull request. All authors of the commits have signed the CLA. 👍
| 阶段 | 任务名 | 状态 | 详情 |
|---|---|---|---|
| 恶意代码检查 | Antipoison | ✅ COMPLETED | >>> |
| 编码安全与规范检查 | Only_doc_commit | ✅ COMPLETED | >>> |
| pre-commit | ✅ COMPLETED | >>> | |
| 开源片段检查 | SCA | ✅ COMPLETED | >>> |
| 开发者测试 | UT_MindSpeed_1 | ✅ COMPLETED | >>> |
| UT_MindSpeed_2 | ✅ COMPLETED | >>> | |
| UT_MindSpeed_3 | ✅ COMPLETED | >>> | |
| UT_MindSpeed_4 | ✅ COMPLETED | >>> | |
| 流水线 | PR-pipeline_MindSpeed-Ops | ✅ COMPLETED | >>> |
- compile : 运行流水线
- retry : 重试流水线所有失败子任务
- retry <任务名> : 仅重试指定失败子任务
- stop : 停止流水线
开发者你好，参照下面的标准再补充对应的信息：
（1）ATK的精度和性能通过截图
（2）UT测试用例通过的截图
（3）triton算子在NPU和GPU上的性能对比
（4）triton算子在NPU和GPU上的精度对比（同数据输入场景）或GPU上验证pytorch小算子与GPU 开源triton算子的精度对比（以验证pytorch小算子标杆准确性）
新增
bwd_dadt_cumsum_fused_kernel算子，用于 Mamba 模型反向传播中 ddt 的融合计算。ut：
atk：
三、910B实测性能
四、A100等效性能折算
结论
使用示例
from mindspeed_ops.api.triton.bwd_dadt_cumsum_fused import bwd_dadt_cumsum_fused # 输入张量: [B, H, S] ddt_out = bwd_dadt_cumsum_fused(ddA_cs, ddA_cs_rev, dA_cs, dA_cs_rev, ddt_out, chunk_size=64)