Describe the current behavior / 问题描述 (Mandatory / 必填)

1. PR 14365 | [Docs] 修正 MoeDistributeDispatchV3/CombineV3 专家均分公式分母
   https://gitcode.com/cann/ops-transformer/merge_requests/14365

2. PR 14363 | [Docs] 更正 SparseFlashMla README 中 cmp_topk_length 的支持条件
   https://gitcode.com/cann/ops-transformer/merge_requests/14363

3. PR 14362 | [Documentation|文档反馈]: BlockSparseAttention 文档未注明 BSND 仅限 Ascend 950
   https://gitcode.com/cann/ops-transformer/merge_requests/14362

4. PR 14358 | [Docs] 更正 GroupedMatmulSwigluQuantV2 文档中的 A8W8 分组参数
   https://gitcode.com/cann/ops-transformer/merge_requests/14358

5. PR 14357 | [Docs] 修正 mamba2_rmsnormgated README 中的 PyTorch 调用名
   https://gitcode.com/cann/ops-transformer/merge_requests/14357

6. PR 14356 | [Docs] 修正 IndexerQuantCache README 的量化粒度与 HiFloat8 scale 说明
   https://gitcode.com/cann/ops-transformer/merge_requests/14356

7. PR 14355 | 修正 GroupedMatmulAdd 的 yRef 参数方向说明
   https://gitcode.com/cann/ops-transformer/merge_requests/14355

8. PR 14353 | [Docs] 修正 GroupedMatmulSwigluQuantV2 MXFP4 dtype 表
   https://gitcode.com/cann/ops-transformer/merge_requests/14353

9. PR 14352 | [Docs] 修正 BlockAttentionResiduals Torch API 的 hidden size 约束
   https://gitcode.com/cann/ops-transformer/merge_requests/14352

10. PR 14350 | [Docs] 补充 BlockAttnResPrepare 的 Ascend 350 支持说明
    https://gitcode.com/cann/ops-transformer/merge_requests/14350

11. PR 14347 | [Docs] 修正 Ascend 950 支持列表中的 MlaProlog 系列 ACLNN 标记
    https://gitcode.com/cann/ops-transformer/merge_requests/14347

12. PR 14346 | [Documentation|文档反馈]: SparseFlashAttentionV2 的 blockTable shape 未说明向上取整容量
    https://gitcode.com/cann/ops-transformer/merge_requests/14346

13. PR 14345 | 修正 MoeDistributeDispatchV3 README 中 ccl_buffer_size 的类型说明
    https://gitcode.com/cann/ops-transformer/merge_requests/14345

14. PR 14342 | [Docs] 修正 MoE Dispatch 共享专家输出容量公式
    https://gitcode.com/cann/ops-transformer/merge_requests/14342

15. PR 14340 | 重复：FlashAttentionVarLenScoreV4 softmax 输出 shape 修正
    https://gitcode.com/cann/ops-transformer/merge_requests/14340

16. PR 14339 | [Documentation|文档反馈]: FlashAttentionVarLenScoreV4 文档未区分 softmax 输出布局对应的 shape
    https://gitcode.com/cann/ops-transformer/merge_requests/14339

17. PR 14338 | [Docs] Clarify kvPaddingSize support in IncreFlashAttention V4
    https://gitcode.com/cann/ops-transformer/merge_requests/14338

18. PR 14337 | [重复，保留 !14336] [Documentation|文档反馈]: MatmulAllReduce README 将 K=0 空内维标为不支持
    https://gitcode.com/cann/ops-transformer/merge_requests/14337

19. PR 14336 | [Documentation|文档反馈]: MatmulAllReduce README 将 K=0 空内维标为不支持
    https://gitcode.com/cann/ops-transformer/merge_requests/14336

20. PR 14333 | [Duplicate] FlashAttentionScore offset 文档更新，请跟进 !14332
    https://gitcode.com/cann/ops-transformer/merge_requests/14333

21. PR 14332 | [Docs] 补充 FlashAttentionScore offset 的随机字节单位和 Philox 对齐规则
    https://gitcode.com/cann/ops-transformer/merge_requests/14332

22. PR 14331 | [Bug-Report|缺陷反馈]: 补充 RecurrentKDA 的头数范围约束
    https://gitcode.com/cann/ops-transformer/merge_requests/14331

23. PR 14329 | [Docs] 修正 GroupedMatmul V4 sparse groupListType 约束描述
    https://gitcode.com/cann/ops-transformer/merge_requests/14329

24. PR 14325 | [Docs] 补充 RingAttentionUpdate README 的 Ascend 950PR/950DT 支持型号
    https://gitcode.com/cann/ops-transformer/merge_requests/14325

25. PR 14286 | 修正 aclnnQuantSparseFlashMlaMetadata 的 cmpResidualKvOptional 必传条件
    https://gitcode.com/cann/ops-transformer/merge_requests/14286

26. PR 14284 | 修正 MhcSinkhornBackward 空 Tensor 支持说明
    https://gitcode.com/cann/ops-transformer/merge_requests/14284

27. PR 14283 | MoeDistributeCombineAddRmsNorm: 修正 expand_x shape 注释
    https://gitcode.com/cann/ops-transformer/merge_requests/14283

28. PR 14282 | 修正 MoeDistributeDispatchV3 README 的输入 dtype
    https://gitcode.com/cann/ops-transformer/merge_requests/14282

29. PR 14281 | 修正 MoeDistributeCombine 文档中的共享专家缩放公式
    https://gitcode.com/cann/ops-transformer/merge_requests/14281

30. PR 14279 | 修正 DistributeBarrier 文档中 worldSize 的数据类型
    https://gitcode.com/cann/ops-transformer/merge_requests/14279

31. PR 14278 | 修正 MoeGatingTopKSoftmax 文档中的 rowIdx 公式和 k 下界
    https://gitcode.com/cann/ops-transformer/merge_requests/14278

32. PR 14260 | [Docs] 修正 RecurrentGatedDeltaRule 公式中的矩阵乘法方向
    https://gitcode.com/cann/ops-transformer/merge_requests/14260

33. PR 14259 | 修正 MoeTokenUnpermuteWithRoutingMapGrad 概率梯度求和范围
    https://gitcode.com/cann/ops-transformer/merge_requests/14259

34. PR 14258 | 更正 MoeTokenPermuteWithRoutingMapGrad README 的混合梯度 dtype 约束
    https://gitcode.com/cann/ops-transformer/merge_requests/14258

35. PR 14257 | 更正 MoeTokenPermuteWithRoutingMapGrad README 的混合梯度 dtype 约束
    https://gitcode.com/cann/ops-transformer/merge_requests/14257

36. PR 14256 | 修正 PoolKeyIndexer Proto 中的 block_size 上限
    https://gitcode.com/cann/ops-transformer/merge_requests/14256

37. PR 14255 | 修正 MoeDistributeDispatch expertTokenNumsType 默认值
    https://gitcode.com/cann/ops-transformer/merge_requests/14255

38. PR 14254 | 修正 MoeInitRoutingQuantV2 动态量化公式
    https://gitcode.com/cann/ops-transformer/merge_requests/14254

39. PR 14253 | 修正 MoeInitRoutingV2 Proto 注释中的 flag 语义
    https://gitcode.com/cann/ops-transformer/merge_requests/14253

40. PR 14250 | 修正 CausalConv1dUpdate 的投机 token 数上界
    https://gitcode.com/cann/ops-transformer/merge_requests/14250

41. PR 14249 | [Docs] 更正 GroupedMatmulSwigluQuant README 中 bias 参数说明
    https://gitcode.com/cann/ops-transformer/merge_requests/14249

42. PR 14246 | fix: 修正 MoeDistributeDispatchSetup 的 comm_cmd_info shape 说明
    https://gitcode.com/cann/ops-transformer/merge_requests/14246

43. PR 14242 | docs(ops-transformer): 修正 --pkg 多 SoC 构建说明
    https://gitcode.com/cann/ops-transformer/merge_requests/14242

44. PR 14241 | [Documentation|文档反馈]: 修正 QuantSparseFlashMla README 中 metadata 必传说明
    https://gitcode.com/cann/ops-transformer/merge_requests/14241

45. PR 14240 | docs(ops-transformer): 修正 MoeInitRoutingV3 FP8 PerBlock expanded_scale shape 公式
    https://gitcode.com/cann/ops-transformer/merge_requests/14240

46. PR 14236 | [Documentation|文档反馈]: 修正 AttentionToFFN shape 列表参数的数据类型说明
    https://gitcode.com/cann/ops-transformer/merge_requests/14236

47. PR 14220 | 修正 MoeGatingTopKSoftmax README 中的 Atlas 推理支持标记
    https://gitcode.com/cann/ops-transformer/merge_requests/14220

48. PR 14217 | docs(ops-transformer): 修正 AlltoAllvGroupedMatMul 的 permute_out 行数注释
    https://gitcode.com/cann/ops-transformer/merge_requests/14217

49. PR 14213 | [Documentation|文档反馈]: 修正 FusedInferAttentionScore PagedAttention 的 actualSeqLengthsKv 长度说明
    https://gitcode.com/cann/ops-transformer/merge_requests/14213

50. PR 14211 | 【update】PromptFlashAttention文档补充FLOAT16 attenMask支持说明
    https://gitcode.com/cann/ops-transformer/merge_requests/14211

51. PR 14208 | docs(ops-transformer): 修正 QuantMatmulAllReduceV3 的 scale rank 文档标注
    https://gitcode.com/cann/ops-transformer/merge_requests/14208

52. PR 14207 | 修正 MatmulSwiglu 高维输入的输出形状说明
    https://gitcode.com/cann/ops-transformer/merge_requests/14207

53. PR 14205 | [Documentation|文档反馈]: GroupedMatmul 可选输出参数补充 nullptr 限制
    https://gitcode.com/cann/ops-transformer/merge_requests/14205

54. PR 14204 | 修正 MatmulAllReduce 的 is_trans_a 属性说明
    https://gitcode.com/cann/ops-transformer/merge_requests/14204

55. PR 14203 | 修正 QkvRmsNormRopeCache 文档中的 K/V 中间输出 shape
    https://gitcode.com/cann/ops-transformer/merge_requests/14203

56. PR 14202 | 修正 MatmulAllReduce 的 310P 支持状态
    https://gitcode.com/cann/ops-transformer/merge_requests/14202

57. PR 14086 | 修复 GenericBlockSparseAttentionGrad 文档中的 N1/N2 约束
    https://gitcode.com/cann/ops-transformer/merge_requests/14086

58. PR 14068 | 补充 CausalConv1d 的 Ascend 350 支持说明
    https://gitcode.com/cann/ops-transformer/merge_requests/14068

59. PR 14066 | 修正 NsaCompressAttention README 中 attentionOut 输出形状说明
    https://gitcode.com/cann/ops-transformer/merge_requests/14066

60. PR 14063 | 修正 MoeDistributeDispatch MX dynamicScalesOut shape 公式
    https://gitcode.com/cann/ops-transformer/merge_requests/14063

61. PR 14062 | 修正 MoeReRouting 的偏移公式
    https://gitcode.com/cann/ops-transformer/merge_requests/14062

62. PR 14061 | 修正 MatmulReduceScatterV2 的 MXFP scale shape 注释
    https://gitcode.com/cann/ops-transformer/merge_requests/14061

63. PR 14058 | 修正 mamba2_causal_conv1d README 权重形状说明
    https://gitcode.com/cann/ops-transformer/merge_requests/14058

64. PR 14056 | 修正 MoeGatingTopKSoftmaxV2 文档中的 k 取值范围
    https://gitcode.com/cann/ops-transformer/merge_requests/14056

65. PR 14055 | Correct GroupedMatmul input dtype documentation
    https://gitcode.com/cann/ops-transformer/merge_requests/14055

66. PR 14052 | 修正 MatmulReduceScatter 输出的通信类型描述
    https://gitcode.com/cann/ops-transformer/merge_requests/14052

67. PR 14051 | [Docs] 明确 ApplyRotaryPosEmb 在 Ascend 950 上的部分 D 约束
    https://gitcode.com/cann/ops-transformer/merge_requests/14051

68. PR 14050 | 修正 SparseFlashMlaSoftmaxL1Norm C 接口文档
    https://gitcode.com/cann/ops-transformer/merge_requests/14050

69. PR 14049 | 修正自定义算子包安装说明中的 vendor 路径
    https://gitcode.com/cann/ops-transformer/merge_requests/14049

70. PR 14048 | [Docs] 修正 QuantAllReduce 返回 shape 文案
    https://gitcode.com/cann/ops-transformer/merge_requests/14048

71. PR 14047 | MatmulReduceScatter Atlas A3 rank_size 文档补充 32 卡
    https://gitcode.com/cann/ops-transformer/merge_requests/14047

72. PR 14044 | Clarify outputOffset support in GroupedMatmulSwigluQuant README
    https://gitcode.com/cann/ops-transformer/merge_requests/14044

73. PR 14042 | 修正 AttentionWorkerCombine 公式中的共享专家项
    https://gitcode.com/cann/ops-transformer/merge_requests/14042

74. PR 14040 | 修正新算子开发指南中的重复实现注册
    https://gitcode.com/cann/ops-transformer/merge_requests/14040

75. PR 14039 | docs(flash-attention-score): 补全 TND inputLayout 说明
    https://gitcode.com/cann/ops-transformer/merge_requests/14039

76. PR 14036 | docs: 修正 AddExample 快速启动命令的工作目录
    https://gitcode.com/cann/ops-transformer/merge_requests/14036

77. PR 13998 | 补充 GroupedMatmul V5 稀疏 groupList 场景说明
    https://gitcode.com/cann/ops-transformer/merge_requests/13998

78. PR 13994 | docs: 修正 BlockSparseAttention 文档中的空 mask 语义
    https://gitcode.com/cann/ops-transformer/merge_requests/13994

79. PR 13926 | 修正 MoeTokenUnpermuteWithEpGrad 文档中的输出属性
    https://gitcode.com/cann/ops-transformer/merge_requests/13926

80. PR 13925 | docs: 修正 ChunkGatedDeltaRule 可选输入参数标注
    https://gitcode.com/cann/ops-transformer/merge_requests/13925

81. PR 13921 | 修正 NormRopeConcat 归一化公式的求和范围
    https://gitcode.com/cann/ops-transformer/merge_requests/13921

82. PR 13824 | 更正 RainFusionAttention 的 BNSD 输入维数说明
    https://gitcode.com/cann/ops-transformer/merge_requests/13824

83. PR 13823 | [Documentation|文档反馈]: 修正 GatherPaKvCache 文档中的 seqLens 取值范围
    https://gitcode.com/cann/ops-transformer/merge_requests/13823

84. PR 13815 | [Docs] 修正 NsaCompress 文档中的 output shape
    https://gitcode.com/cann/ops-transformer/merge_requests/13815

85. PR 13804 | AttentionWorkerScheduler README 补充调用约束
    https://gitcode.com/cann/ops-transformer/merge_requests/13804

86. PR 13794 | FusedCausalConv1dV2 文档修正 APC cacheIndices 首维形状
    https://gitcode.com/cann/ops-transformer/merge_requests/13794

87. PR 13702 | 修正 MoeDistributeCombineV2 README 常量专家输入 shape
    https://gitcode.com/cann/ops-transformer/merge_requests/13702

88. PR 13701 | 修正 MoeInitRoutingQuant 的 expanded_x 形状说明
    https://gitcode.com/cann/ops-transformer/merge_requests/13701

89. PR 13700 | 修正 MoeInitRoutingQuant 的 expanded_x 形状说明
    https://gitcode.com/cann/ops-transformer/merge_requests/13700

90. PR 13699 | 补充 AttentionToFFN README 中可选输入的使用约束
    https://gitcode.com/cann/ops-transformer/merge_requests/13699

91. PR 13696 | 修正 FfnWorkerBatching FP8/FP4 token_dtype 的平台支持范围
    https://gitcode.com/cann/ops-transformer/merge_requests/13696

92. PR 13677 | [Docs] 修正 RotaryStride README 的参数契约
    https://gitcode.com/cann/ops-transformer/merge_requests/13677

93. PR 13607 | [Docs] 补充 MoeGatingTopKSoftmaxV2 可选输入说明
    https://gitcode.com/cann/ops-transformer/merge_requests/13607

94. PR 13606 | 修正 QuantGroupedMatmulInplaceAdd 的 scale1 参数说明
    https://gitcode.com/cann/ops-transformer/merge_requests/13606

95. PR 13601 | 修正 RingAttentionUpdate 对 softmax max 的取值说明
    https://gitcode.com/cann/ops-transformer/merge_requests/13601

96. PR 13600 | 修正 RecurrentGatedDeltaRule 文档中的可选参数说明
    https://gitcode.com/cann/ops-transformer/merge_requests/13600

97. PR 13002 | 修复 GroupedMatmulFinalizeRouting README 中失效的样例链接
    https://gitcode.com/cann/ops-transformer/merge_requests/13002

Environment Information / 环境信息 (Mandatory / 必填)

1

Steps to reproduce the issue / 重现步骤 (Mandatory / 必填)

1

Describe the expected behavior / 预期结果 (Mandatory / 必填)

1

Related log / screenshot / 日志 / 截图 (Mandatory / 必填)

1

Special notes for this issue/备注 (Optional / 选填)