pta a5反合
Co-authored-by: yanhf<yanhaifeng5@huawei.com>
# message auto-generated for no-merge-commit merge:
!3568 merge master into master
pta a5反合
Created-by: yanhf
Commit-by: yanhf
Merged-by: ascend-robot
Description: <!-- Thanks for sending a pull request!
-->
**What type of PR is this?**
> Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line:
>
> /kind bug
> /kind task
> /kind feature
**What does this PR do / why do we need it**:
**Special notes for your reviewers**:
See merge request: Ascend/op-plugin!3568
OP hash supports thread-level
Co-authored-by: wang-guangbin<wgb_strive@163.com>
# message auto-generated for no-merge-commit merge:
!3681 merge atb2 into master
OP hash supports thread-level
Created-by: wang-guangbin
Commit-by: wang-guangbin
Merged-by: ascend-robot
Description: <!-- Thanks for sending a pull request!
-->
**What type of PR is this?**
> Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line:
>
> /kind bug
> /kind task
> /kind feature
**What does this PR do / why do we need it**:
op仓的atb op复用采用线程级单例。
当前op仓的atb缓存设计为进程级别的单列,多线程场景存在op复用冲突,atb底层的setup和exector不支持相同的op在不同线程。
**Special notes for your reviewers**:
See merge request: Ascend/op-plugin!3681
Fix potential integer overflow issues in npu_paged_cache_load
Co-authored-by: wang-guangbin<wgb_strive@163.com>
# message auto-generated for no-merge-commit merge:
!3413 merge bugfix into master
Fix potential integer overflow issues in npu_paged_cache_load
Created-by: wang-guangbin
Commit-by: wang-guangbin
Merged-by: ascend-robot
Description: <!-- Thanks for sending a pull request!
-->
**What type of PR is this?**
> Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line:
>
> /kind bug
> /kind task
> /kind feature
**What does this PR do / why do we need it**:
**Special notes for your reviewers**:
See merge request: Ascend/op-plugin!3413
OP hash supports thread-level
Co-authored-by: wang-guangbin<wgb_strive@163.com>
# message auto-generated for no-merge-commit merge:
!3681 merge atb2 into master
OP hash supports thread-level
Created-by: wang-guangbin
Commit-by: wang-guangbin
Merged-by: ascend-robot
Description: <!-- Thanks for sending a pull request!
-->
**What type of PR is this?**
> Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line:
>
> /kind bug
> /kind task
> /kind feature
**What does this PR do / why do we need it**:
op仓的atb op复用采用线程级单例。
当前op仓的atb缓存设计为进程级别的单列,多线程场景存在op复用冲突,atb底层的setup和exector不支持相同的op在不同线程。
**Special notes for your reviewers**:
See merge request: Ascend/op-plugin!3681
OP hash supports thread-level
Co-authored-by: wang-guangbin<wgb_strive@163.com>
# message auto-generated for no-merge-commit merge:
!3681 merge atb2 into master
OP hash supports thread-level
Created-by: wang-guangbin
Commit-by: wang-guangbin
Merged-by: ascend-robot
Description: <!-- Thanks for sending a pull request!
-->
**What type of PR is this?**
> Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line:
>
> /kind bug
> /kind task
> /kind feature
**What does this PR do / why do we need it**:
op仓的atb op复用采用线程级单例。
当前op仓的atb缓存设计为进程级别的单列,多线程场景存在op复用冲突,atb底层的setup和exector不支持相同的op在不同线程。
**Special notes for your reviewers**:
See merge request: Ascend/op-plugin!3681
rename 910_95
Co-authored-by: MrMC-<shiqunze@h-partners.com>
# message auto-generated for no-merge-commit merge:
!4108 merge master-re950 into master
rename 910_95
Created-by: MrMC-
Commit-by: MrMC-
Merged-by: ascend-robot
Description: <!-- Thanks for sending a pull request!
-->
**What type of PR is this?**
> Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line:
>
> /kind bug
> /kind task
> /kind feature
**What does this PR do / why do we need it**:
**Special notes for your reviewers**:
See merge request: Ascend/op-plugin!4108
rename 910_95
Co-authored-by: MrMC-<shiqunze@h-partners.com>
# message auto-generated for no-merge-commit merge:
!4108 merge master-re950 into master
rename 910_95
Created-by: MrMC-
Commit-by: MrMC-
Merged-by: ascend-robot
Description: <!-- Thanks for sending a pull request!
-->
**What type of PR is this?**
> Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line:
>
> /kind bug
> /kind task
> /kind feature
**What does this PR do / why do we need it**:
**Special notes for your reviewers**:
See merge request: Ascend/op-plugin!4108
Add handling logic for y2Out shape.
Co-authored-by: 花无懿<liuxuehui4@huawei.com>
# message auto-generated for no-merge-commit merge:
!3760 merge master into master
Add handling logic for y2Out shape.
Created-by: huawuyi
Commit-by: 花无懿
Merged-by: ascend-robot
Description: <!-- Thanks for sending a pull request!
-->
**What type of PR is this?**
> Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line:
>
> /kind bug
> /kind task
> /kind feature
**What does this PR do / why do we need it**:
**Special notes for your reviewers**:
See merge request: Ascend/op-plugin!3760
add aaqmm api and fix bugs
Co-authored-by: wangkechen<wangkechen3@huawei.com>
# message auto-generated for no-merge-commit merge:
!4150 merge aaqmm2 into master
add aaqmm api and fix bugs
Created-by: Kiana1216
Commit-by: wangkechen
Merged-by: ascend-robot
Description: <!-- Thanks for sending a pull request!
-->
**What type of PR is this?**
> Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line:
>
> /kind bug
> /kind task
> /kind feature
**What does this PR do / why do we need it**:
**Special notes for your reviewers**:
See merge request: Ascend/op-plugin!4150
add aaqmm api and fix bugs
Co-authored-by: wangkechen<wangkechen3@huawei.com>
# message auto-generated for no-merge-commit merge:
!4150 merge aaqmm2 into master
add aaqmm api and fix bugs
Created-by: Kiana1216
Commit-by: wangkechen
Merged-by: ascend-robot
Description: <!-- Thanks for sending a pull request!
-->
**What type of PR is this?**
> Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line:
>
> /kind bug
> /kind task
> /kind feature
**What does this PR do / why do we need it**:
**Special notes for your reviewers**:
See merge request: Ascend/op-plugin!4150
attention update support fp16 bf16
Co-authored-by: qiumingli<liqiuming4@huawei.com>
# message auto-generated for no-merge-commit merge:
!3636 merge master into master
attention update support fp16 bf16
Created-by: qiumingli
Commit-by: qiumingli
Merged-by: ascend-robot
Description: <!-- Thanks for sending a pull request!
-->
**What type of PR is this?**
> Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line:
>
> /kind bug
> /kind task
> /kind feature
**What does this PR do / why do we need it**:
**Special notes for your reviewers**:
See merge request: Ascend/op-plugin!3636
add npu_clipped_swiglu
Co-authored-by: jzj007<jiangzhijie9@huawei.com>
# message auto-generated for no-merge-commit merge:
!3487 merge cs_without_md into master
add npu_clipped_swiglu
Created-by: jzj007
Commit-by: jzj007
Merged-by: ascend-robot
Description: <!-- Thanks for sending a pull request!
-->
**What type of PR is this?**
> Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line:
>
> /kind feature
**What does this PR do / why do we need it**:
add npu_clipped_swiglu
**Special notes for your reviewers**:
See merge request: Ascend/op-plugin!3487
新增npu_dual_level_quant_matmul
Co-authored-by: chaoying-zhang<zhangchaoying@huawei.com>
# message auto-generated for no-merge-commit merge:
!4158 merge dlqbmm_a4w4 into master
新增npu_dual_level_quant_matmul
Created-by: chaoying-zhang
Commit-by: chaoying-zhang
Merged-by: ascend-robot
Description: <!-- Thanks for sending a pull request!
-->
**What type of PR is this?**
> Uncomment only one /kind <> line, hit enter to put that in a new line, and remove leading whitespaces from that line:
>
> /kind bug
> /kind task
/kind feature
**What does this PR do / why do we need it**:
新增npu_dual_level_quant_matmul算子,实现MXFP4二级量化,补偿精度
**Special notes for your reviewers**:
See merge request: Ascend/op-plugin!4158