xuchx
/
mindspore
mirror of https://gitee.com/mindspore/mindspore.git
0
0
Fork 1
Code Issues Packages Projects Releases Wiki Activity
101,501 Commits 227 Branches 58 Tags 5.4 GiB
Commit Graph

366 Commits

Author SHA1 Message Date
wujueying 1823b93a72 add nsa_compress_attention 2025-11-30 13:23:03 +08:00
wang_ziqi dc5e1af6a0 mindspore support nsa_select_attention & nsa_select_attention_grad 
------

Signed-off-by: wang_ziqi <wangziqi4@huawei.com>
 2025-10-20 14:11:10 +08:00
liangchenghui ba5fa7b598 Add ops.nsa_compress ops. 2025-10-16 11:38:39 +08:00
huan d0a4053e66 modify api order 2025-07-15 19:05:28 +08:00
lilinjie e1752b769f adapt RingAttentionUpdate 2025-06-30 16:24:39 +08:00
mengxian f465d81656 mindspore.ops 2025-04-07 19:44:44 +08:00
yuhan 2796266337 modify format 2025-03-28 16:47:00 +08:00
liuchuting 8da304548b Fix ops.unbind/ops.unique/ops.unique_consecutive/ops.unique_with_pad docs. 2025-03-27 20:46:36 +08:00
王振邦 0791f3b8a4 add move_to chinese doc 2025-03-14 10:35:05 +08:00
lilinjie d11f10d434 adapt aclnnMoeTokenPermute 2025-02-18 10:05:57 +08:00
wang_ziqi 37114d0f49 Add ops.speed_fusion_attention 
------

Signed-off-by: wang_ziqi <wangziqi4@huawei.com>
 2025-02-17 14:30:31 +08:00
shenhaojing 307eb4574c add moe_tokens_unpermute 2025-02-13 19:56:45 +08:00
wutengda cf1491c501 ops: Add fused_infer_attention_score() functional api 
Add fused_infer_attention_score functional api to align with the
aclnn inferface `aclnnFusedInferAttentionScoreV2`. This functional
api supports pynative, kbk, ge and semi auto parallel mode, where
ge and semi auto parallel mode are implemented by reusing the
existing primitive `FusedInferAttentionScore`, with some minor
adaptations made for the new parameters.

To align with PyTorch NPU, we made the following adjustments to the
parameters of the functional interface, which have some slight differences
from the FusedInferAttentionScore primitive:

  * attn_mask renamed to atten_mask
  * scale_value renamed to scale
  * Input parameter order adjustment:
    * key_antiquant_scale/offset and value_antiquant_scale/offset are
      moved forward by two positions, placed after antiquant_offset.
    * key/value_antiquant_mode is moved forward by one position, placed
      after antiquant_mode.
  * key/value data types: The key and value parameters now support the
    Tensor type and are no longer restricted to being passed as arrays.
 2025-02-11 21:24:19 +08:00
yanghaoran d49a28a018
回退 'Pull Request !81071 : add ops.moe_token_unpermute' 2025-02-11 12:53:55 +00:00
yanghaoran 171acceaa5
回退 'Pull Request !81159 : ops: Add fused_infer_attention_score() functional api' 2025-02-11 12:24:20 +00:00
i-robot 494353ec54
!81159 ops: Add fused_infer_attention_score() functional api 
Merge pull request !81159 from stavewu/dev-fia-v9
 2025-02-11 10:40:14 +00:00
wutengda 45cad4dd57 ops: Add fused_infer_attention_score() functional api 
Add fused_infer_attention_score functional api to align with the
aclnn inferface `aclnnFusedInferAttentionScoreV2`. This functional
api supports pynative, kbk, ge and semi auto parallel mode, where
ge and semi auto parallel mode are implemented by reusing the
existing primitive `FusedInferAttentionScore`, with some minor
adaptations made for the new parameters.

To align with PyTorch NPU, we made the following adjustments to the
parameters of the functional interface, which have some slight differences
from the FusedInferAttentionScore primitive:

  * attn_mask renamed to atten_mask
  * scale_value renamed to scale
  * Input parameter order adjustment:
    * key_antiquant_scale/offset and value_antiquant_scale/offset are
      moved forward by two positions, placed after antiquant_offset.
    * key/value_antiquant_mode is moved forward by one position, placed
      after antiquant_mode.
  * key/value data types: The key and value parameters now support the
    Tensor type and are no longer restricted to being passed as arrays.
 2025-02-11 09:42:34 +08:00
shenhaojing 66a14dc084 token unpermute 2025-02-11 09:18:45 +08:00
lilinjie 279c8ac6b8 add doc for ops.flash_attention_score(aka FAS) 2025-02-05 14:19:28 +08:00
LiYouKang 6aacee820d 新增注册列表 2025-01-21 17:12:35 +08:00
yuhan 4be7a7c785 modify ops.rst display and DynamicProfilerMonitor members 2024-12-26 14:49:22 +08:00
fengyihang 99875fc610 fix ops rst 2024-12-20 15:50:05 +08:00
lilinjie 4acfd47f7d release IFA doc 2024-12-09 17:30:08 +08:00
cycychenyi c5fb0f37d3 add all_gather_matmul and matmul_reduce_scatter 2024-12-04 23:22:12 +08:00
lilinjie 5293dd7a60 release ops.prompt_flash_attention doc 2024-11-21 15:03:14 +08:00
xiaopeng a201bcf546 fix docs 2024-11-04 17:18:57 +08:00
yuhan 77565a7bd6 modify urls 2024-10-14 10:44:01 +08:00
qiuzhongya 46db10efe1 add docs for rotary_position_embedding 
add docs for add layer norm
 2024-09-21 07:25:00 +08:00
MengXY107 d9ae73e120 tensordump on master 
fix tensordump doc

fix import
 2024-09-05 07:08:56 -04:00
wujueying 07421d52a0 ops.cast add docs 2024-08-29 14:27:51 +08:00
XianglongZeng f8384b2812 add rms_norm aclnn api to master 2024-06-22 19:25:14 +08:00
dingjinshan 5f484bd6d7 aclnnLayerNorm 2024-04-26 16:11:41 +08:00
zhang_xu_hao1230 ef2d7bcb19 add aclnn for onehot 2024-04-26 16:11:17 +08:00
ckey_Dou cf76a9cd3a support embedding 2024-04-26 16:10:06 +08:00
huan e6f4d8afe0 modify the error links 2024-04-18 10:11:22 +08:00
i-robot 236b24d57c
!67921 modify url r2.3 
Merge pull request !67921 from 俞涵/code_docs_r23q1
 2024-04-07 00:37:00 +00:00
z00841464 8b0b8409ab delete fft numpy/scipy api list 2024-04-03 14:15:07 +08:00
yuhan cabc2d5b1f modify url r2.3 2024-04-03 10:07:37 +08:00
huangle02 51be056f1e add rfft doc 2024-03-12 15:00:53 +08:00
z00841464 9a3511280d add fft/ifft/fft2/ifft2/fftn/ifftn docs 2024-03-11 16:22:02 +08:00
z00841464 6243eb0d4b add fftshift/ifftshift at cpu/aicpu backend with dyn shape 2024-01-17 21:53:11 +08:00
huan f50e2a3c94 add contents in rst files 2.3 2023-12-14 14:17:24 +08:00
huan b257cffe86 add contents in files 2.2 2023-12-11 11:21:03 +08:00
huan 1128ccd423 modify the order in rst file 2023-10-13 17:38:03 +08:00
shenzhangyi 513dcf1b27 Add Eps 2023-08-10 15:27:41 +08:00
i-robot a7b4edce80
!56615 add tensor eq 
Merge pull request !56615 from Henry Shi/tensor_eq
 2023-08-03 07:27:35 +00:00
lilinjie cff3e85e8c del Eig doc from website 2023-08-03 10:15:34 +08:00
Henry ccf46a0f3c add tensor and ops eq 2023-08-02 23:08:58 +08:00
fengyihang 7b5a26f884 clip by norm 2023-07-19 17:32:44 +08:00
yide12 0d1c2915d3 add_nanmean_nanmedian_slice_select_scatter_master 2023-06-30 18:11:29 +08:00