op-plugin/test/test_custom_ops · mihudan/op-plugin

ascend-robot[Feature][Master] Support npu_dense_lightning_indexer_softmax_lse API

文件	最后提交记录	最后更新时间
test_all_gather_base_mm.py	Add npu_all_gather_base_mm testcase Co-authored-by: wang-guangbin<wgb_strive@163.com> # message auto-generated for no-merge-commit merge: !3339 merge allgather into master Add npu_all_gather_base_mm testcase 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: 本地验证通过： ![image.png](https://raw.gitcode.com/user-images/assets/7403085/a54d8e40-6542-4484-b006-eeb700b306ff/image.png 'image.png') See merge request: Ascend/op-plugin!3339	11 个月前
test_all_gather_dtype_fault.py	check output dtype when input fp16/bf16 Co-authored-by: hanhaocheng<hanhaocheng2@huawei.com> # message auto-generated for no-merge-commit merge: !3635 merge master into master check output dtype when input fp16/bf16 Created-by: HertzHan Commit-by: hanhaocheng 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 What does this PR do / why do we need it: add output datatype check so that it keep the same with input Special notes for your reviewers: See merge request: Ascend/op-plugin!3635	9 个月前
test_all_gather_quant_mm.py	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	10 个月前
test_alltoallv_gmm.py	!2448 ADD ALLTOALLVGMM AND GMMALLTOALLV Merge pull request !2448 from 梅文瑄/master	1 年前
test_atb_flash_attention.py	!2185 Added the atb operator monitoring case. Merge pull request !2185 from 史佳馨/master	1 年前
test_atb_flash_attention_prefix_v2.py	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	9 个月前
test_atb_flash_attention_qlens.py	!2229 The self-attention PREFIX_ENCODER and pagedattention_qlens atb operators are added. Merge pull request !2229 from 史佳馨/master	1 年前
test_atb_flash_attention_unpad.py	!2185 Added the atb operator monitoring case. Merge pull request !2185 from 史佳馨/master	1 年前
test_atb_flash_attention_v2.py	!2578 Added the alibi scenario of the atb pa operator, alibi scenario of the… Merge pull request !2578 from 史佳馨/master	1 年前
test_atb_group_topk.py	!2175 Add group_topk. Merge pull request !2175 from 王夏夏/master	1 年前
test_atb_linear.py	!2733 Fix cpu tensor deadlocks Merge pull request !2733 from wgb/atb	1 年前
test_atb_mla_int8_nzcache.py	!2632 add atb mla and prefix encoder ut Merge pull request !2632 from wgb/atb	1 年前
test_atb_mla_krope_ctkv.py	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	9 个月前
test_atb_mla_with_lse.py	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	9 个月前
test_atb_mlapo.py	!2646 Add atb meta and doc feature and mlapo adaption Merge pull request !2646 from wgb/atb	1 年前
test_atb_paged_attention.py	PA surpport aclgraph Co-authored-by: wang-guangbin<wgb_strive@163.com> # message auto-generated for no-merge-commit merge: merge atb into master PA surpport aclgraph 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: 1. ATB，C++接入方式下发方式新增workpspace支持外部传入，以支持aclgraph的update特性。 2. 新增特性使能atb pa接口，支持aclgraph Special notes for your reviewers: See merge request: Ascend/op-plugin!3204	1 年前
test_atb_paged_attention_mla.py	!2456 modify atb monitoring case Merge pull request !2456 from 史佳馨/master	1 年前
test_atb_paged_attention_quant.py	!2456 modify atb monitoring case Merge pull request !2456 from 史佳馨/master	1 年前
test_atb_paged_attention_splitfuse.py	!2373 Fix libatb.so not found in some docker Merge pull request !2373 from wgb/bugfix	1 年前
test_atb_paged_attention_v2.py	PA surpport aclgraph Co-authored-by: wang-guangbin<wgb_strive@163.com> # message auto-generated for no-merge-commit merge: merge atb into master PA surpport aclgraph 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: 1. ATB，C++接入方式下发方式新增workpspace支持外部传入，以支持aclgraph的update特性。 2. 新增特性使能atb pa接口，支持aclgraph Special notes for your reviewers: See merge request: Ascend/op-plugin!3204	1 年前
test_atb_paged_cache_load.py	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	10 个月前
test_atb_quant_per_tensor.py	!2560 Atb support aclgraph Merge pull request !2560 from wgb/atb_nz	1 年前
test_atb_reshape_and_cache.py	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	9 个月前
test_atb_reshape_and_cache_siso.py	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	9 个月前
test_atb_ring_mla.py	!2625 The ut care of npu_ring_mla and npu_fused_add_topk_div is added. Merge pull request !2625 from 史佳馨/master	1 年前
test_atb_rope.py	!2617 ATB support torch.compile with fullgraph Merge pull request !2617 from wgb/atb3	1 年前
test_atb_selfattention_prefix_encoder.py	!2632 add atb mla and prefix encoder ut Merge pull request !2632 from wgb/atb	1 年前
test_attention_to_ffn.py	【op-plugin】分离算子attention_to_ffn/ffn_to_attention PTA 适配 Co-authored-by: ykyro01<caozhuoyi@huawei.com> # message auto-generated for no-merge-commit merge: !3831 merge master into master 【op-plugin】分离算子attention_to_ffn/ffn_to_attention PTA 适配 Created-by: ykyro01 Commit-by: ykyro01 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!3831	9 个月前
test_attention_worker_schedule.py	适配支持两类算子：ffn_worker_scheduler_/ffn_worker_scheduler、 attention_worker_scheduler_/attention_worker_scheduler Co-authored-by: panweicheng<panweicheng@huawei.com> # message auto-generated for no-merge-commit merge: !3397 merge master into master 适配支持两类算子：ffn_worker_scheduler_/ffn_worker_scheduler、 attention_worker_scheduler_/attention_worker_scheduler Created-by: pwc10490 Commit-by: panweicheng Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> What type of PR is this? > /kind feature What does this PR do / why do we need it: 适配支持两类算子： torch_npu.attention_worker_scheduler torch_npu.attention_worker_scheduler_（inplace算子） torch_npu.ffn_worker_scheduler torch_npu.ffn_worker_scheduler_（inplace算子） Special notes for your reviewers: See merge request: Ascend/op-plugin!3397	11 个月前
test_distribute_barrier.py	barrier Co-authored-by: davon<liuwenda4@huawei.com> # message auto-generated for no-merge-commit merge: !3372 merge master into master barrier Created-by: Davon14272 Commit-by: davon 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: _npu_distribute_barrier adds two optional inputs: time_out and elastic_info Special notes for your reviewers: See merge request: Ascend/op-plugin!3372	11 个月前
test_ffn_to_attention.py	【op-plugin】分离算子attention_to_ffn/ffn_to_attention PTA 适配 Co-authored-by: ykyro01<caozhuoyi@huawei.com> # message auto-generated for no-merge-commit merge: !3831 merge master into master 【op-plugin】分离算子attention_to_ffn/ffn_to_attention PTA 适配 Created-by: ykyro01 Commit-by: ykyro01 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!3831	9 个月前
test_ffn_worker_schedule.py	适配支持两类算子：ffn_worker_scheduler_/ffn_worker_scheduler、 attention_worker_scheduler_/attention_worker_scheduler Co-authored-by: panweicheng<panweicheng@huawei.com> # message auto-generated for no-merge-commit merge: !3397 merge master into master 适配支持两类算子：ffn_worker_scheduler_/ffn_worker_scheduler、 attention_worker_scheduler_/attention_worker_scheduler Created-by: pwc10490 Commit-by: panweicheng Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> What type of PR is this? > /kind feature What does this PR do / why do we need it: 适配支持两类算子： torch_npu.attention_worker_scheduler torch_npu.attention_worker_scheduler_（inplace算子） torch_npu.ffn_worker_scheduler torch_npu.ffn_worker_scheduler_（inplace算子） Special notes for your reviewers: See merge request: Ascend/op-plugin!3397	11 个月前
test_fused_add_topk_div.py	!3072 npu_fused_add_topk_div benchmark function modification Merge pull request !3072 from 史佳馨/master	1 年前
test_gmm_alltoallv.py	!2448 ADD ALLTOALLVGMM AND GMMALLTOALLV Merge pull request !2448 from 梅文瑄/master	1 年前
test_group_norm_silu.py	【950】op-plugin回合主仓：moe_re_routing/moe_init_routing_v2/scaled_masked_softmax/group_norm_silu/moe_gating_top_k_softmax_v2 Co-authored-by: xujiachen8<xujiachen8@h-partners.com> # message auto-generated for no-merge-commit merge: !3429 merge vector-prs into master 【950】op-plugin回合主仓：moe_re_routing/moe_init_routing_v2/scaled_masked_softmax/group_norm_silu/moe_gating_top_k_softmax_v2 Created-by: xujiachen8 Commit-by: xujiachen8 Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> What type of PR is this? op-plugin回合主仓： 1) npu_group_norm_silu支持混合精度 (部分平台) 2) npu_moe_re_routing 支持float8 (部分平台) 3) npu_moe_init_routing_v2 修复参数active_expert_range导致的空指针异常 4) npu_scaled_masked_softmax 支持路径5 5) npu_moe_gating_top_k_softmax_v2 支持PTA What does this PR do / why do we need it: Special notes for your reviewers: See merge request: Ascend/op-plugin!3429	10 个月前
test_hans_compress.py	!2385 add HansEncode ans HansDeocde op api Merge pull request !2385 from Kingsleyandher/master	1 年前
test_incre_flash_attention.py	!2100 fa bugfix Merge pull request !2100 from realmadrid1016/fa_bugfix	1 年前
test_masked_softmax_with_rel_pos_bias.py	!1200 Fix 910B skip ut use decorator Merge pull request !1200 from 王广斌/master_cherry	2 年前
test_mla_prolog.py	!2711 Modify MlaProlog operator UT Merge pull request !2711 from 刘宇/master	1 年前
test_mla_prolog_v2.py	!2711 Modify MlaProlog operator UT Merge pull request !2711 from 刘宇/master	1 年前
test_mla_prolog_v3.py	feature:mlaprologv3 mxfp8 ut Co-authored-by: zuoyw<wangxu350@huawei.com> # message auto-generated for no-merge-commit merge: !3826 merge master into master feature:mlaprologv3 mxfp8 ut Created-by: zuoyw Commit-by: zuoyw 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!3826	9 个月前
test_mm_all_reduce_base.py	!1988 Fix some UTs on atlas_a2 part 3 Merge pull request !1988 from 张向龙3/fix_910b_ut_3	1 年前
test_mm_all_reduce_base_add.py	!1988 Fix some UTs on atlas_a2 part 3 Merge pull request !1988 from 张向龙3/fix_910b_ut_3	1 年前
test_mm_all_reduce_base_perblock.py	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	10 个月前
test_mm_reduce_scatter_base.py	npu_mm_reduce_scatter_base support quant and comm_mode aiv Co-authored-by: wang-guangbin<wgb_strive@163.com> # message auto-generated for no-merge-commit merge: merge mm_rs into master npu_mm_reduce_scatter_base support quant and comm_mode aiv 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: 本地验证通过 ![image.png](https://raw.gitcode.com/user-images/assets/7403085/3b23aa82-4d92-4ae0-9617-b48ea4f056e1/image.png 'image.png') See merge request: Ascend/op-plugin!3194	1 年前
test_moe_distribute_combine.py	!2838 revert change of moe_distribute_dispatch/combine v1 interface Merge pull request !2838 from sincatter/master_dis_v1	1 年前
test_moe_distribute_combineA2.py	mc2 dispatch&combine v2 add aclnnV4 for master Co-authored-by: yyoean<1114146082@qq.com> # message auto-generated for no-merge-commit merge: !3538 merge master into master mc2 dispatch&combine v2 add aclnnV4 for master Created-by: yyoean Commit-by: yyoean 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: https://gitcode.com/Ascend/op-plugin/issues/25 See merge request: Ascend/op-plugin!3538	10 个月前
test_moe_distribute_combine_add_rms_norm.py	!3062 DispatchV3/CombineV3/CombineaddrmsnormV2 update Merge pull request !3062 from Oblivion/master	1 年前
test_moe_distribute_combine_v2.py	!3062 DispatchV3/CombineV3/CombineaddrmsnormV2 update Merge pull request !3062 from Oblivion/master	1 年前
test_moe_distribute_dispatch.py	!2838 revert change of moe_distribute_dispatch/combine v1 interface Merge pull request !2838 from sincatter/master_dis_v1	1 年前
test_moe_distribute_dispatchA2.py	mc2 dispatch&combine v2 add aclnnV4 for master Co-authored-by: yyoean<1114146082@qq.com> # message auto-generated for no-merge-commit merge: !3538 merge master into master mc2 dispatch&combine v2 add aclnnV4 for master Created-by: yyoean Commit-by: yyoean 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: https://gitcode.com/Ascend/op-plugin/issues/25 See merge request: Ascend/op-plugin!3538	10 个月前
test_moe_distribute_dispatch_v2.py	!3062 DispatchV3/CombineV3/CombineaddrmsnormV2 update Merge pull request !3062 from Oblivion/master	1 年前
test_moe_update_expert.py	!2954 npu_moe_eplb_update_expert PTA Merge pull request !2954 from Oblivion/master	1 年前
test_npu_add_layer_norm.py	!1502 AddLayerNorm return FP32 when input different dtype Merge pull request !1502 from 何曾庚/hzg_mix	2 年前
test_npu_add_layer_norm_backward.py	【950】DSQ and ARNQ add attr&BNGTU and BNR support opapi&LN positive and negative operator performance optimization Co-authored-by: cai-chengchao<caichengchao1@h-partners.com> # message auto-generated for no-merge-commit merge: !3432 merge A5_back_master into master 【950】DSQ and ARNQ add attr&BNGTU and BNR support opapi&LN positive and negative operator performance optimization Created-by: cai-chengchao Commit-by: cai-chengchao 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!3432	10 个月前
test_npu_add_rms_norm.py	!1200 Fix 910B skip ut use decorator Merge pull request !1200 from 王广斌/master_cherry	2 年前
test_npu_add_rms_norm_cast.py	!2438 add new add_rms_norm_cast Merge pull request !2438 from liuyun/master	1 年前
test_npu_add_rms_norm_dynamic_quant.py	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	9 个月前
test_npu_add_rms_norm_quant.py	【950】DSQ and ARNQ add attr&BNGTU and BNR support opapi&LN positive and negative operator performance optimization Co-authored-by: cai-chengchao<caichengchao1@h-partners.com> # message auto-generated for no-merge-commit merge: !3432 merge A5_back_master into master 【950】DSQ and ARNQ add attr&BNGTU and BNR support opapi&LN positive and negative operator performance optimization Created-by: cai-chengchao Commit-by: cai-chengchao 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!3432	10 个月前
test_npu_add_rms_norm_v2.py	[feat]add AddRmsNormInplace Co-authored-by: cao-ren-lei<caorenlei@huawei.com> # message auto-generated for no-merge-commit merge: !3558 merge master into master [feat]add AddRmsNormInplace Created-by: caorenlei Commit-by: cao-ren-lei 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: 增加aclnnInplaceAddRmsNorm接口对应的pta接口 Special notes for your reviewers: See merge request: Ascend/op-plugin!3558	9 个月前
test_npu_addmm.py	!1935 Fix some UTs on 910B Merge pull request !1935 from 张向龙3/fix_ut_910b	1 年前
test_npu_advance_step_flashattn.py	!3031 add AdvanceStep to support speculative inference Merge pull request !3031 from 陈赵旻熠/AdvanceStep	1 年前
test_npu_all_to_all_matmul.py	alltoallmm and mmalltoall pta Co-authored-by: wangkechen<wangkechen3@huawei.com> # message auto-generated for no-merge-commit merge: !3922 merge pta into master alltoallmm and mmalltoall pta 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!3922	9 个月前
test_npu_apply_adam_w.py	接口回归主线 Co-authored-by: 鲜磊<xianlei4@h-partners.com> # message auto-generated for no-merge-commit merge: !3573 merge conv into master 接口回归主线 Created-by: xianlei2 Commit-by: 鲜磊 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!3573	9 个月前
test_npu_apply_rotary_pos_emb.py	support tnd format Co-authored-by: ghost<zhengwenbang@huawei.com> # message auto-generated for no-merge-commit merge: !3355 merge master into master support tnd format Created-by: weixin_44564637 Commit-by: ghost 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!3355	11 个月前
test_npu_attention_update.py	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	9 个月前
test_npu_attention_worker_combine.py	Add AttentionWorkerCombine PTA Co-authored-by: weixin_36491311<lingwanxin@huawei.com> # message auto-generated for no-merge-commit merge: merge master into master Add AttentionWorkerCombine PTA Created-by: weixin_36491311 Commit-by: weixin_36491311 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!3176	1 年前
test_npu_attn_softmax.py	!2335 Add npu_attn_softmax. Merge pull request !2335 from 王夏夏/master	1 年前
test_npu_batch_gather_matmul.py	!2454 remove ut skip Merge pull request !2454 from 沈珈靓/master	1 年前
test_npu_clipped_swiglu.py	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	10 个月前
test_npu_convert_weight_to_int4pack.py	npu_convert_weight_to_int4pack支持3维ND Co-authored-by: chaoying-zhang<zhangchaoying@huawei.com> # message auto-generated for no-merge-commit merge: !3628 merge gmm_a16w4 into master npu_convert_weight_to_int4pack支持3维ND 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 AR20250604177532 What does this PR do / why do we need it: GMM A16W4需要npu_convert_weight_to_int4pack支持3维ND，当前实现输出数据固定是2维，pack结果错误，需要修复 Special notes for your reviewers: See merge request: Ascend/op-plugin!3628	10 个月前
test_npu_cross_entropy_loss.py	!2150 [Feature] Add npu_cross_entropy_loss. Merge pull request !2150 from 刘嘉巍/master	1 年前
test_npu_deep_norm.py	!1200 Fix 910B skip ut use decorator Merge pull request !1200 from 王广斌/master_cherry	2 年前
test_npu_deep_norm_backward.py	!1200 Fix 910B skip ut use decorator Merge pull request !1200 from 王广斌/master_cherry	2 年前
test_npu_dense_lightning_indexer_softmax_lse.py	[Feature][Master] Support npu_dense_lightning_indexer_softmax_lse API Co-authored-by: chengpeng25<chengpeng9@huawei.com> # message auto-generated for no-merge-commit merge: !4059 merge dev_das_master into master [Feature][Master] Support npu_dense_lightning_indexer_softmax_lse API Created-by: chengpeng25 Commit-by: chengpeng25 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: Support DSA(dense) API: dense_indexer_lightning_softmax_lse. test pass: ![image.png](https://raw.gitcode.com/user-images/assets/7403085/63c5b0d5-412f-44e9-abc6-e5a2fe4cdcd0/image.png 'image.png') Special notes for your reviewers: See merge request: Ascend/op-plugin!4059	8 个月前
test_npu_dequant_swiglu_quant.py	【950】DSQ and ARNQ add attr&BNGTU and BNR support opapi&LN positive and negative operator performance optimization Co-authored-by: cai-chengchao<caichengchao1@h-partners.com> # message auto-generated for no-merge-commit merge: !3432 merge A5_back_master into master 【950】DSQ and ARNQ add attr&BNGTU and BNR support opapi&LN positive and negative operator performance optimization Created-by: cai-chengchao Commit-by: cai-chengchao 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!3432	10 个月前
test_npu_dropout_gen_mask.py	!1600 add opapi of dropout_gen_mask Merge pull request !1600 from 赖长铃/custom_dropout_aclnn	2 年前
test_npu_dynamic_block_quant.py	!3119 [Feat] Add npu_dynamic_block_quant docs and meta Merge pull request !3119 from zhangqiongwen/master_npu_dynamic_block_quant_doc	1 年前
test_npu_dynamic_mx_quant_with_dual_axis.py	DynamicMxQuantWithDualAxis Co-authored-by: yin-peng-hw<yinpeng37@huawei.com> # message auto-generated for no-merge-commit merge: !3788 merge DynamicMxQuantWithDualAxis into master DynamicMxQuantWithDualAxis Created-by: yin-peng Commit-by: yin-peng-hw 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!3788	9 个月前
test_npu_dynamic_quant.py	new commit for DynamicQuantV2(perchannel mode) Co-authored-by: dyd-dmy<dengyidan@huawei.com> # message auto-generated for no-merge-commit merge: !3794 merge DynamicQuant into master new commit for DynamicQuantV2(perchannel mode) Created-by: dyd-dmy Commit-by: dyd-dmy Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> Co-authored-by: 李龙千禧<lilongqianxi@h-partners.com> Co-authored-by: 周亮<zhouliang88@huawei.com> What type of PR is this? > /kind feature What does this PR do / why do we need it: 在DynamicQuant算子基础上新增perchannel场景 Special notes for your reviewers: See merge request: Ascend/op-plugin!3794	9 个月前
test_npu_dynamic_quant_asymmetric.py	new commit for DynamicQuantV2(perchannel mode) Co-authored-by: dyd-dmy<dengyidan@huawei.com> # message auto-generated for no-merge-commit merge: !3794 merge DynamicQuant into master new commit for DynamicQuantV2(perchannel mode) Created-by: dyd-dmy Commit-by: dyd-dmy Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> Co-authored-by: 李龙千禧<lilongqianxi@h-partners.com> Co-authored-by: 周亮<zhouliang88@huawei.com> What type of PR is this? > /kind feature What does this PR do / why do we need it: 在DynamicQuant算子基础上新增perchannel场景 Special notes for your reviewers: See merge request: Ascend/op-plugin!3794	9 个月前
test_npu_exponential.py	!1953 Prevent lambda input from being negative Merge pull request !1953 from zzx/master	1 年前
test_npu_ffn.py	!2075 ffn ut update Merge pull request !2075 from zqh/1231	1 年前
test_npu_ffn_worker_batching.py	add FfnWorkerBatching Co-authored-by: 安冬<andong2@huawei.com> # message auto-generated for no-merge-commit merge: merge master into master add FfnWorkerBatching Created-by: andong_hw Commit-by: 安冬 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!3179	1 年前
test_npu_flash_attention.py	remove keep_prob cast Co-authored-by: caihualilili<liangfuzhan@h-partners.com> # message auto-generated for no-merge-commit merge: merge keep_prob_remove_cast into master remove keep_prob cast Created-by: caihualilili Commit-by: caihualilili 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!3200	11 个月前
test_npu_flash_attention_grad.py	remove keep_prob cast Co-authored-by: caihualilili<liangfuzhan@h-partners.com> # message auto-generated for no-merge-commit merge: merge keep_prob_remove_cast into master remove keep_prob cast Created-by: caihualilili Commit-by: caihualilili 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!3200	11 个月前
test_npu_flash_attention_grad_v2.py	[feat]npu_fusion_attention_v2 adapt for fav5 Co-authored-by: zqwenn<zhangqiongwen@huawei.com> # message auto-generated for no-merge-commit merge: !3526 merge master_fav5_adpt into master [feat]npu_fusion_attention_v2 adapt for fav5 Created-by: zqwenn Commit-by: zqwenn 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: npu_fusion_attention_v2新增softmax_layout和sink入参新增场景测试用例下一个pr提交 See merge request: Ascend/op-plugin!3526	10 个月前
test_npu_flash_attention_v2.py	add npu_quant_fusion_attention and npu_quant_fusion_attention_grad Co-authored-by: fangchao_24<fangchao24@hisilicon.com> # message auto-generated for no-merge-commit merge: !3854 merge master into master add npu_quant_fusion_attention and npu_quant_fusion_attention_grad Created-by: fangchao_24 Commit-by: fangchao_24 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: Special notes for your reviewers: See merge request: Ascend/op-plugin!3854	9 个月前
test_npu_fused_floyd_attention.py	Modify the backward ut passes of npu_fused_floyd_attention. Co-authored-by: shi-jiaxin9<shijiaxin10@h-partners.com> # message auto-generated for no-merge-commit merge: !3885 merge master into master Modify the backward ut passes of npu_fused_floyd_attention. Created-by: shi-jiaxin9 Commit-by: shi-jiaxin9 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 bugModify the backward ut passes of npu_fused_floyd_attention. What does this PR do / why do we need it: Modify the backward ut passes of npu_fused_floyd_attention. Special notes for your reviewers: See merge request: Ascend/op-plugin!3885	9 个月前
test_npu_fused_infer_attention_score.py	【950】 FusedInferAttentionScore 相关 Co-authored-by: xyangjia<jiaxuyang1@h-partners.com> # message auto-generated for no-merge-commit merge: !3473 merge new into master 【950】 FusedInferAttentionScore 相关 Created-by: xyangjia Commit-by: xyangjia 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!3473	10 个月前
test_npu_fused_infer_attention_score_v2.py	【950】 FusedInferAttentionScore 相关 Co-authored-by: xyangjia<jiaxuyang1@h-partners.com> # message auto-generated for no-merge-commit merge: !3473 merge new into master 【950】 FusedInferAttentionScore 相关 Created-by: xyangjia Commit-by: xyangjia 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!3473	10 个月前
test_npu_fused_infer_attention_score_v2_with_workspace.py	support npu_fused_infer_attention_score_v2 tiling update in NPUGraph Co-authored-by: fengye<vinsmoke.feng@huawei.com> # message auto-generated for no-merge-commit merge: !3924 merge master into master support npu_fused_infer_attention_score_v2 tiling update in NPUGraph Created-by: fengye0903 Commit-by: fengye 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: 用例执行结果 #@unittest.skip( # "Skipping due to outdated CANN version; please update CANN to the latest version and remove this skip") #@SupportedDevices(['Ascend910B']) def test_npu_fused_infer_attention_score_v2_ifa_return_lse_return_softmax_lse_true_with_workspace(self, device="npu"): query = torch.randn(1, 32, 1, 128, dtype=torch.float16).npu() key = torch.randn(1, 32, 2048, 128, dtype=torch.float16).npu() value = torch.randn(1, 32, 2048, 128, dtype=torch.float16).npu() head_dim = 128 return_softmax_lse = True supported_output, lse_out = self.supported_op_exec(query, key, value, head_dim, 1, 32, 1, return_softmax_lse) custom_output = self.custom_op_exec_with_workspace(query, key, value, head_dim, True) attention_output = custom_output[0] softmaxlse_output = custom_output[1] if __name__ == "__main__": run_tests() [root@devserver-hps-feb34b1c-00033 f00941739]# python ./op-plugin/test/test_custom_ops/test_npu_fused_infer_attention_score_v2_with_workspace.py ss. ---------------------------------------------------------------------- Ran 3 tests in 2.006s OK (skipped=2) 整网测试结果： [root@devserver-hps-feb34b1c-00033 omni-npu-2]# curl http://7.150.9.163:8001/v1/completions -H "Content-Type: application/json" -d '{"prompt":"请介绍一下苏州，不少于100字，请用中文回复","model":"Pangu72B","temperature":0.6,"max_tokens":120}' {"id":"cmpl-94a127a570be5f14","object":"text_completion","created":1767012795,"model":"Pangu72B","choices":[{"index":0,"text":"。\n苏州，位于中国江苏省东南部，是一座拥有2500多年历史的古城，以其独特的园林艺术和丰富的文化遗产而闻名。苏州园林是中国古典园林的代表，被誉为“园林之城”，其中拙政园、留园等都是著名的景点，展现了精巧的设计和深远的文化内涵。此外，苏州还以丝绸制造业著称，有“丝绸之府”的美誉。苏州的评弹、昆曲等传统艺术形式也体现了这座城市深厚的文化底蕴。今天的苏州，不仅保留了丰富的历史遗产，还是一个现代化发展的城市，经济繁荣，科技和教育水平先进。","logprobs":null,"finish_reason":"stop","stop_reason":45892,"token_ids":null,"prompt_logprobs":null,"prompt_token_ids":null}],"service_tier":null,"system_fingerprint":null,"us age":{"prompt_tokens":12,"total_tokens":131,"completion_tokens":119,"prompt_tokens_details":null},"kv_transfer_params":null}[root@devserver-hps-feb34b1c-00033 omni-npu-2]# [root@devserver-hps-feb34b1c-00033 omni-npu-2]# See merge request: Ascend/op-plugin!3924	9 个月前
test_npu_fused_infer_attention_score_with_workspace.py	add _npu_fused_infer_attention_score_infer_output Co-authored-by: yanpengquan<yanpengquan@huawei.com> # message auto-generated for no-merge-commit merge: !3891 merge master into master add _npu_fused_infer_attention_score_infer_output Created-by: yanpengquan Commit-by: yanpengquan 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!3891	9 个月前
test_npu_fused_matmul.py	[950] npu_fuesed_matmul接口新增 & npu_grouped_matmul_add_接口新增参数 Co-authored-by: eternityk<taoyikun@huawei.com> # message auto-generated for no-merge-commit merge: !3446 merge master into master [950] npu_fuesed_matmul接口新增 & npu_grouped_matmul_add_接口新增参数 Created-by: eternityk Commit-by: eternityk Merged-by: ascend-robot Description: 【新功能提交】功能描述 npu_fuesed_matmul接口新增 & npu_grouped_matmul_add_接口新增参数相关Issue https://gitee.com/ascend/op-plugin-for-lingqu2.0/commit/5afab9f02e512b1f38362364e2ac2a3b60a9659a?w=1 https://gitee.com/ascend/op-plugin-for-lingqu2.0/commit/426ed212a59383739cbd19bb245aa4858fbbfed7?w=1 https://gitee.com/ascend/op-plugin-for-lingqu2.0/pulls/161/files?w=1 https://gitee.com/ascend/op-plugin-for-lingqu2.0/pulls/175/files?w=1 https://gitee.com/ascend/op-plugin-for-lingqu2.0/pulls/180/files?w=1 Checklist [x] 符合编程规范&&代码红线 [x] 具备UT测试用例看护&&测试用例为有效用例（若无测试用例看护请说明原因） [x] 是否涉及对外接口变更是 [x] 是否涉及官网文档修改不涉及文档修改，A5资料暂不开放 See merge request: Ascend/op-plugin!3446	10 个月前
test_npu_gather_pa_kv_cache.py	【950】新增npu_gather_pa_kv_cache算子相关内容 Co-authored-by: gaojie120<gaojie120@huawei.com> # message auto-generated for no-merge-commit merge: !3433 merge master into master 【950】新增npu_gather_pa_kv_cache算子相关内容 Created-by: gaojie120 Commit-by: gaojie120 Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> What type of PR is this? 新增 test/test_custom_ops/test_npu_gather_pa_kv_cache.py op_plugin/ops/opapi/GatherPaKvCacheKernelNpuOpApi.cpp 修改 op_plugin/config op_plugin/python/meta op_plugin/ops/opapi/GatherPaKvCache What does this PR do / why do we need it: Special notes for your reviewers: See merge request: Ascend/op-plugin!3433	10 个月前
test_npu_gather_sparse_index.py	!3085 [Feat] Add npu_gather_sparse_index_backward adpt Merge pull request !3085 from zhangqiongwen/master_npu_gather_sparse_index_backward	1 年前
test_npu_gelu.py	!1965 GeluV2/GeluGradV2 PTA接口与入图功能代码修改 Merge pull request !1965 from 禹广飞/geluv2_pta	1 年前
test_npu_gelu_backward.py	!1965 GeluV2/GeluGradV2 PTA接口与入图功能代码修改 Merge pull request !1965 from 禹广飞/geluv2_pta	1 年前
test_npu_gelu_mul.py	Added support for the npu_gelu_mul operator. Co-authored-by: shi-jiaxin9<shijiaxin10@h-partners.com> # message auto-generated for no-merge-commit merge: !3418 merge master into master Added support for the npu_gelu_mul operator. Created-by: shi-jiaxin9 Commit-by: shi-jiaxin9 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: Added support for the npu_gelu_mul operator. Special notes for your reviewers: See merge request: Ascend/op-plugin!3418	10 个月前
test_npu_gelu_quant.py	【950】cast && quant类算子回合 Co-authored-by: 李根<ligen75@h-partners.com> Co-authored-by: Wang_Xing001<wangxing73@huawei.com> Co-authored-by: lilongqianxi<lilongqianxi@h-partners.com> Co-authored-by: 田野<tianye82@huawei.com> Co-authored-by: 季骏<jijun1@huawei.com> Co-authored-by: luo-yongqing<luoyongqing4@h-partners.com> Co-authored-by: 赵珩<zhaoheng32@h-partners.com> # message auto-generated for no-merge-commit merge: !3431 merge pta_plugin_sync into master 【950】cast && quant类算子回合 Created-by: yuanbin_22 Commit-by: 赵珩;luo-yongqing;季骏;lilongqianxi;Wang_Xing001;田野;李根 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: A5代码回合主线： 1）npu_dtype_cast支持float4和float8数据类型 2）新增npu_gelu_quant，npu_grouped_dynamic_mx_quant，npu_dynamic_mx_quant，npu_dynamic_block_quant接口 3）QuantUpdateScatter 图模式设置输出dtype，以适配hifp8 4）【Quantize】输出y设置 desc dtype Special notes for your reviewers: See merge request: Ascend/op-plugin!3431	10 个月前
test_npu_gemma_rms_norm.py	为npu_gemma_rms_norm算子提供meta注册 Co-authored-by: panweicheng<panweicheng@huawei.com> # message auto-generated for no-merge-commit merge: !4014 merge master into master 为npu_gemma_rms_norm算子提供meta注册 Created-by: pwc10490 Commit-by: panweicheng 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!4014	8 个月前
test_npu_group_norm_swish.py	!2125 add group_norm_swish and group_norm_swish_grad Merge pull request !2125 from 詹昊/groupnormswish	1 年前
test_npu_group_quant.py	!1839 add new op group_quant Merge pull request !1839 from 田野/dev_groupquant	1 年前
test_npu_grouped_matmul.py	support wqbmmv2 and qbmmv4 pta nz single op Co-authored-by: 张泉新<zhangquanxin7@h-partners.com> # message auto-generated for no-merge-commit merge: !3844 merge support_wqbmmv2_pta_nz_1203 into master support wqbmmv2 and qbmmv4 pta nz single op Created-by: zhangquanxin Commit-by: 张泉新 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!3844	9 个月前
test_npu_grouped_matmul_add.py	update meta function Co-authored-by: Zichun Ye<zichun.ye@huawei.com> # message auto-generated for no-merge-commit merge: !3778 merge meta_update_master into master update meta function Created-by: zichun_ye Commit-by: Zichun Ye 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: add meta function Special notes for your reviewers: See merge request: Ascend/op-plugin!3778	9 个月前
test_npu_grouped_matmul_finalize_routing.py	GmmFR 支持mxfp8,mxfp4 Co-authored-by: lirh<lironghui4@huawei.com> # message auto-generated for no-merge-commit merge: !3901 merge gmmfr_1225 into master GmmFR 支持mxfp8,mxfp4 Created-by: lironghui4 Commit-by: lirh 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!3901	9 个月前
test_npu_grouped_matmul_swiglu_quant.py	!2630 reshapeandcache support nz and dtype int8 Merge pull request !2630 from 史佳馨/master	1 年前
test_npu_grouped_matmul_swiglu_quant_v2.py	torch-fp4 Co-authored-by: hyhhh14<huangyuehua6@huawei.com> # message auto-generated for no-merge-commit merge: !3907 merge merge-torch into master torch-fp4 Created-by: hyhhh14 Commit-by: hyhhh14 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!3907	9 个月前
test_npu_kronecker_quant.py	!2243 npu_kronecker_quant添加可选参数clip_ratio Merge pull request !2243 from 陈海杰/master	1 年前
test_npu_kv_quant_sparse_flash_attention.py	LI\QLI\SFA\QSFA PTA适配 Co-authored-by: zzzyh22<zhangyaohui22@huawei.com> # message auto-generated for no-merge-commit merge: !3425 merge master into master LI\QLI\SFA\QSFA PTA适配 Created-by: zzzyh22 Commit-by: zzzyh22 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!3425	10 个月前
test_npu_kv_rms_norm_rope_cache.py	修改格式 Co-authored-by: 15084323657<xulin91@huawei.com> # message auto-generated for no-merge-commit merge: !3457 merge kv_pr_v2 into master 修改格式 Created-by: Will_09 Commit-by: 15084323657 Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> 算子新增v2功能实现： kv_rms_norm_rope_cache_支持kv分离（新增可选输入tensor v）和ROPE(RMS(Dk))计算 tiling,kernel侧，增加非对称量化和Norm下量化功能。实现输入kv N维度扩展 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: Special notes for your reviewers: See merge request: Ascend/op-plugin!3457	10 个月前
test_npu_layer_norm_eval.py	!1304 Bugfix: fix the problem of optional parameter in npu_layer_norm_eval Merge pull request !1304 from 王广斌/master_cherry	2 年前
test_npu_lightning_indexer.py	LI\QLI\SFA\QSFA PTA适配 Co-authored-by: zzzyh22<zhangyaohui22@huawei.com> # message auto-generated for no-merge-commit merge: !3425 merge master into master LI\QLI\SFA\QSFA PTA适配 Created-by: zzzyh22 Commit-by: zzzyh22 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!3425	10 个月前
test_npu_matmul_all_to_all.py	alltoallmm and mmalltoall pta Co-authored-by: wangkechen<wangkechen3@huawei.com> # message auto-generated for no-merge-commit merge: !3922 merge pta into master alltoallmm and mmalltoall pta 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!3922	9 个月前
test_npu_moe_compute_tokens.py	!1433 add moe compute tokens PTA Merge pull request !1433 from 冯佳伟/test	2 年前
test_npu_moe_finalize_routing_v2.py	!2081 moe_finalize_routing_v2 适配支持 drop_pad_mode 场景 Merge pull request !2081 from langlh/moefix	1 年前
test_npu_moe_gating_top_k_softmax.py	!1935 Fix some UTs on 910B Merge pull request !1935 from 张向龙3/fix_ut_910b	1 年前
test_npu_moe_gating_top_k_softmax_v2.py	【950】op-plugin回合主仓：moe_re_routing/moe_init_routing_v2/scaled_masked_softmax/group_norm_silu/moe_gating_top_k_softmax_v2 Co-authored-by: xujiachen8<xujiachen8@h-partners.com> # message auto-generated for no-merge-commit merge: !3429 merge vector-prs into master 【950】op-plugin回合主仓：moe_re_routing/moe_init_routing_v2/scaled_masked_softmax/group_norm_silu/moe_gating_top_k_softmax_v2 Created-by: xujiachen8 Commit-by: xujiachen8 Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> What type of PR is this? op-plugin回合主仓： 1) npu_group_norm_silu支持混合精度 (部分平台) 2) npu_moe_re_routing 支持float8 (部分平台) 3) npu_moe_init_routing_v2 修复参数active_expert_range导致的空指针异常 4) npu_scaled_masked_softmax 支持路径5 5) npu_moe_gating_top_k_softmax_v2 支持PTA What does this PR do / why do we need it: Special notes for your reviewers: See merge request: Ascend/op-plugin!3429	10 个月前
test_npu_moe_gating_topk.py	!3061 modify npu_moe_gating_top_k api doc Merge pull request !3061 from 汤PC/master	1 年前
test_npu_moe_init_routing.py	!1537 add moe_init_routing Merge pull request !1537 from 汤PC/master	2 年前
test_npu_moe_init_routing_quant.py	!2930 新增MoeInitRoutingQuant算子接口 Merge pull request !2930 from CY-Slightwind/master_cy	1 年前
test_npu_moe_init_routing_v2.py	MIRV3适配mxfp8量化 Co-authored-by: xujiachen8<xujiachen8@h-partners.com> # message auto-generated for no-merge-commit merge: !3884 merge mirv3_mxfp8 into master MIRV3适配mxfp8量化 Created-by: xujiachen8 Commit-by: xujiachen8 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 /kind feature What does this PR do / why do we need it: MoeInitRoutingV3适配mxfp8量化，新增quant_mode=2、3模式，适配输出shape和dtype的变化 Special notes for your reviewers: See merge request: Ascend/op-plugin!3884	9 个月前
test_npu_moe_token_permute_and_unpermute.py	!2590 add npu_moe_token_permute and npu_moe_token_unpermute Merge pull request !2590 from shaoyf/permute_2	1 年前
test_npu_moe_token_permute_and_unpermute_backward.py	!2983 fix probs check and add grad output test Merge pull request !2983 from 吴尚伦/master	1 年前
test_npu_moe_token_permute_with_routing_map.py	!2871 Add npu_moe_token_permute_with_routing_map and testcase Merge pull request !2871 from wgb/bugfix	1 年前
test_npu_moe_token_unpermute_with_routing_map.py	!2727 [Feature] npu_moe_token_unpermute_with_routing_map Merge pull request !2727 from 刘嘉巍/mha	1 年前
test_npu_nsa_compress.py	!2720 add nsa ut Merge pull request !2720 from 詹昊/master	1 年前
test_npu_nsa_compress_attention.py	!2720 add nsa ut Merge pull request !2720 from 詹昊/master	1 年前
test_npu_nsa_compress_attention_infer.py	!2964 nsa_compress_attention_infer support BSND Merge pull request !2964 from songjianquan/master_nsa	1 年前
test_npu_nsa_compress_grad.py	!2720 add nsa ut Merge pull request !2720 from 詹昊/master	1 年前
test_npu_nsa_compress_infer.py	!2720 add nsa ut Merge pull request !2720 from 詹昊/master	1 年前
test_npu_nsa_select_attention.py	!2720 add nsa ut Merge pull request !2720 from 詹昊/master	1 年前
test_npu_nsa_select_attention_grad.py	!2720 add nsa ut Merge pull request !2720 from 詹昊/master	1 年前
test_npu_nsa_select_attention_infer.py	!2936 NSA SelectAttention infer: add TND support. Merge pull request !2936 from njupt_sjj/master	1 年前
test_npu_qkv_rms_norm_rope_cache.py	新增qkv_rms_norm_rope_cache算子 Co-authored-by: fengqiuyue<fengqiuyue@huawei.com> # message auto-generated for no-merge-commit merge: !4028 merge QKV_NO_DOC into master 新增qkv_rms_norm_rope_cache算子 Created-by: fengqiuyue Commit-by: fengqiuyue 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 qkv_rms_norm_rope_cache Special notes for your reviewers: See merge request: Ascend/op-plugin!4028	8 个月前
test_npu_quant_all_reduce.py	add new op quant_all_reduce pta intf Co-authored-by: yifux<xiongyifu1@huawei.com> # message auto-generated for no-merge-commit merge: !3707 merge pr_quant into master add new op quant_all_reduce pta intf Created-by: xiongyifu Commit-by: yifux 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!3707	9 个月前
test_npu_quant_conv2d.py	【950】conv forward A5 opplugin code Co-authored-by: huafeng793<chenhuafeng2@huawei.com> # message auto-generated for no-merge-commit merge: !3430 merge conv_A5 into master 【950】conv forward A5 opplugin code Created-by: huafeng793 Commit-by: huafeng793 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 feature What does this PR do / why do we need it: npu_quant_conv2d接口，A5新增参数、参数变更，以新增支持hifloat8数据类型。 npu_quant_conv2d接口，A5新增支持opapi接口，支持aclnn Special notes for your reviewers: See merge request: Ascend/op-plugin!3430	10 个月前
test_npu_quant_flash_attention.py	add npu_quant_fusion_attention and npu_quant_fusion_attention_grad Co-authored-by: fangchao_24<fangchao24@hisilicon.com> # message auto-generated for no-merge-commit merge: !3854 merge master into master add npu_quant_fusion_attention and npu_quant_fusion_attention_grad Created-by: fangchao_24 Commit-by: fangchao_24 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: Special notes for your reviewers: See merge request: Ascend/op-plugin!3854	9 个月前
test_npu_quant_flash_attention_grad.py	add npu_quant_fusion_attention and npu_quant_fusion_attention_grad Co-authored-by: fangchao_24<fangchao24@hisilicon.com> # message auto-generated for no-merge-commit merge: !3854 merge master into master add npu_quant_fusion_attention and npu_quant_fusion_attention_grad Created-by: fangchao_24 Commit-by: fangchao_24 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: Special notes for your reviewers: See merge request: Ascend/op-plugin!3854	9 个月前
test_npu_quant_lightning_indexer.py	LI\QLI\SFA\QSFA PTA适配 Co-authored-by: zzzyh22<zhangyaohui22@huawei.com> # message auto-generated for no-merge-commit merge: !3425 merge master into master LI\QLI\SFA\QSFA PTA适配 Created-by: zzzyh22 Commit-by: zzzyh22 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!3425	10 个月前
test_npu_quant_matmul.py	Fixed the error of judging whether to call npu_trans_quant_param in npu_quant_matmul Co-authored-by: liyinghao10<liyinghao10@h-partners.com> # message auto-generated for no-merge-commit merge: !3832 merge qbmm into master Fixed the error of judging whether to call npu_trans_quant_param in npu_quant_matmul Created-by: liyinghao10 Commit-by: liyinghao10 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 在npu_quant_matmul单算子接口中，在scale实际传入的数据类型使FLOAT32的情况下，如果想要转换成UINT64的数据类型，需要调用npu_trans_quant_param接口。在输入为INT8的情况下，输出类型为BFLOAT16时不调用npu_trans_quant_param接口，但是FP8/HIF8的输入类型下，输出为BFLOAT16时仍需调用该接口。本PR为修复此漏洞。 > /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!3832	9 个月前
test_npu_quant_matmul_all_to_all.py	alltoallmm and mmalltoall pta Co-authored-by: wangkechen<wangkechen3@huawei.com> # message auto-generated for no-merge-commit merge: !3922 merge pta into master alltoallmm and mmalltoall pta 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!3922	9 个月前
test_npu_quant_matmul_reduce_sum.py	!2806 add npu_quant_matmul_reduce_sum Merge pull request !2806 from AaronTang/qbmm_reduce_sum	1 年前
test_npu_quant_reduce_scatter.py	add new op quant_reduce_scatter pta intf Co-authored-by: yifux<xiongyifu1@huawei.com> # message auto-generated for no-merge-commit merge: !3745 merge pr_quant into master add new op quant_reduce_scatter pta intf Created-by: xiongyifu Commit-by: yifux 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!3745	9 个月前
test_npu_quant_scatter.py	【950】cast && quant类算子回合 Co-authored-by: 李根<ligen75@h-partners.com> Co-authored-by: Wang_Xing001<wangxing73@huawei.com> Co-authored-by: lilongqianxi<lilongqianxi@h-partners.com> Co-authored-by: 田野<tianye82@huawei.com> Co-authored-by: 季骏<jijun1@huawei.com> Co-authored-by: luo-yongqing<luoyongqing4@h-partners.com> Co-authored-by: 赵珩<zhaoheng32@h-partners.com> # message auto-generated for no-merge-commit merge: !3431 merge pta_plugin_sync into master 【950】cast && quant类算子回合 Created-by: yuanbin_22 Commit-by: 赵珩;luo-yongqing;季骏;lilongqianxi;Wang_Xing001;田野;李根 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: A5代码回合主线： 1）npu_dtype_cast支持float4和float8数据类型 2）新增npu_gelu_quant，npu_grouped_dynamic_mx_quant，npu_dynamic_mx_quant，npu_dynamic_block_quant接口 3）QuantUpdateScatter 图模式设置输出dtype，以适配hifp8 4）【Quantize】输出y设置 desc dtype Special notes for your reviewers: See merge request: Ascend/op-plugin!3431	10 个月前
test_npu_quantize.py	npu_quantize接口input数据格式支持NZ Co-authored-by: wkp<1289706727@qq.com> # message auto-generated for no-merge-commit merge: merge master-quantize into master npu_quantize接口input数据格式支持NZ Created-by: kunpengW Commit-by: wkp 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: ![image.png](https://raw.gitcode.com/user-images/assets/7403085/0984ff44-7fd0-4bd0-aac1-f621b7a6a750/image.png 'image.png') See merge request: Ascend/op-plugin!3219	1 年前
test_npu_recurrent_gated_delta_rule.py	add RGDR op-plugin Co-authored-by: NeverLCY<liuchunyuan@huawei.com> # message auto-generated for no-merge-commit merge: !3476 merge RGDR into master add RGDR op-plugin Created-by: NeverLCY Commit-by: NeverLCY 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!3476	10 个月前
test_npu_rms_norm.py	!1754 RmsNorm y dtype should same as input x dtype Merge pull request !1754 from 何曾庚/hzg_y_dtype	2 年前
test_npu_rms_norm_backward.py	!1935 Fix some UTs on 910B Merge pull request !1935 from 张向龙3/fix_ut_910b	1 年前
test_npu_rms_norm_quant.py	add npu_rms_norm_quant interface again Co-authored-by: lihui488<lihui488@huawei.com> # message auto-generated for no-merge-commit merge: merge develop into master add npu_rms_norm_quant interface again Created-by: lihui488 Commit-by: lihui488 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!3214	1 年前
test_npu_rotary_mul.py	RotaryPositionEmbedding adds option input rotate Co-authored-by: lilening<lilening@huawei.com> # message auto-generated for no-merge-commit merge: !3802 merge master into master RotaryPositionEmbedding adds option input rotate Created-by: lilening Commit-by: lilening Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> What type of PR is this? >/kind feature What does this PR do / why do we need it: RotaryPositionEmbedding adds option input rotate Special notes for your reviewers: UT用例本地已测试通过 ![image.png](https://raw.gitcode.com/user-images/assets/7403085/fc2998e5-a2e7-430e-8b85-54fd9c86c1c3/image.png 'image.png') ![image.png](https://raw.gitcode.com/user-images/assets/7403085/c800c488-6cc6-49c7-b840-abfc120534cd/image.png 'image.png') See merge request: Ascend/op-plugin!3802	9 个月前
test_npu_rotary_mul_backward.py	Added support for three-dimensional TND in both forward and reverse inputs of aclnnRotaryPositionEmbedding. Co-authored-by: shi-jiaxin9<shijiaxin10@h-partners.com> # message auto-generated for no-merge-commit merge: !3549 merge master into master Added support for three-dimensional TND in both forward and reverse inputs of aclnnRotaryPositionEmbedding. Created-by: shi-jiaxin9 Commit-by: shi-jiaxin9 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: Added support for three-dimensional TND in both forward and reverse inputs of aclnnRotaryPositionEmbedding. Special notes for your reviewers: See merge request: Ascend/op-plugin!3549	10 个月前
test_npu_scatter_nd_update.py	!1935 Fix some UTs on 910B Merge pull request !1935 from 张向龙3/fix_ut_910b	1 年前
test_npu_scatter_pa_cache.py	ScatterPaCache Co-authored-by: l30072255<liuhuansheng1@h-partners.com> # message auto-generated for no-merge-commit merge: !3434 merge master into master ScatterPaCache Created-by: fishing_artist Commit-by: l30072255 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!3434	10 个月前
test_npu_sim_exponential_.py	[SR]expensonal precision alignment Co-authored-by: liangsongwei<liangsongwei@huawei.com> # message auto-generated for no-merge-commit merge: !3455 merge master into master [SR]expensonal precision alignment Created-by: liangsongwei Commit-by: liangsongwei 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!3455	10 个月前
test_npu_sparse_flash_attention.py	SFA资料修改/补充sparse_size为0的用例 Co-authored-by: zzzyh22<zhangyaohui22@huawei.com> # message auto-generated for no-merge-commit merge: !3753 merge master into master SFA资料修改/补充sparse_size为0的用例 Created-by: zzzyh22 Commit-by: zzzyh22 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: 补充SFA算子参数描述，对attention_mode，sparse_block_size补充说明；增加SFA异常拦截用例。 Special notes for your reviewers: See merge request: Ascend/op-plugin!3753	9 个月前
test_npu_swi_glu.py	!1946 FIX：修改swi_glu正反向ut测试用例 Merge pull request !1946 from 曹坤/master	1 年前
test_npu_swi_glu_backward.py	!1946 FIX：修改swi_glu正反向ut测试用例 Merge pull request !1946 from 曹坤/master	1 年前
test_npu_swiglu_quant.py	add torch_npu.npu_swiglu_quant api, based on aclnnSwiGluQuantV2 Co-authored-by: zhaoyu20121997<nanzhaogang@qq.com> # message auto-generated for no-merge-commit merge: merge swiglu_pr into master add torch_npu.npu_swiglu_quant api, based on aclnnSwiGluQuantV2 Created-by: zhaoyu20121997 Commit-by: zhaoyu20121997 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: 新增torch_npu.npu_swiglu_quant接口，接入aclnnSwiGluQuantV2算子 Special notes for your reviewers: 接口说明请参考提交的文件docs/context/torch_npu-npu_swiglu_quant.md See merge request: Ascend/op-plugin!3167	1 年前
test_npu_top_k_top_p.py	Modify the Added support for scenarios where p or k is set to None in npu_top_k_top_p. Co-authored-by: shi-jiaxin9<shijiaxin10@h-partners.com> # message auto-generated for no-merge-commit merge: !3344 merge master into master Modify the Added support for scenarios where p or k is set to None in npu_top_k_top_p. Created-by: shi-jiaxin9 Commit-by: shi-jiaxin9 Merged-by: ascend-robot Description: 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: Modify the Added support for scenarios where p or k is set to None in npu_top_k_top_p. Special notes for your reviewers: See merge request: Ascend/op-plugin!3344	11 个月前
test_npu_trans_quant_param.py	!1935 Fix some UTs on 910B Merge pull request !1935 from 张向龙3/fix_ut_910b	1 年前
test_npu_transpose_batchmatmul.py	!2638 修复TBMM算子图融合参数未校验的问题 Merge pull request !2638 from 马鲁胜/DTS_TBMM	1 年前
test_npu_weight_quant_batchmatmul.py	check weight quant matmul nz is available Co-authored-by: 张泉新<zhangquanxin7@h-partners.com> # message auto-generated for no-merge-commit merge: !3664 merge master into master check weight quant matmul nz is available Created-by: zhangquanxin Commit-by: 张泉新 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!3664	8 个月前
test_obfuscation_calculate.py	!3043 Add obfuscation_initialize obfucation_finialize obfuscation_calculate Merge pull request !3043 from 蒋学博/master	1 年前
test_obfuscation_finalize.py	!3043 Add obfuscation_initialize obfucation_finialize obfuscation_calculate Merge pull request !3043 from 蒋学博/master	1 年前
test_obfuscation_initialize.py	!3043 Add obfuscation_initialize obfucation_finialize obfuscation_calculate Merge pull request !3043 from 蒋学博/master	1 年前
test_prompt_flash_attention.py	!2100 fa bugfix Merge pull request !2100 from realmadrid1016/fa_bugfix	1 年前
test_quant_grouped_matmul_inplace_add.py	[950] QuantGroupedMatmulInplaceAdd A5内容合入 Co-authored-by: liyinghao10<liyinghao10@h-partners.com> # message auto-generated for no-merge-commit merge: !3534 merge qgmmia into master [950] QuantGroupedMatmulInplaceAdd A5内容合入 Created-by: liyinghao10 Commit-by: liyinghao10 Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> What type of PR is this? op-plugin回合主仓：新增npu_add_quant_gmm和npu_add_quant_gmm_接口（仅A5） What does this PR do / why do we need it: Special notes for your reviewers: See merge request: Ascend/op-plugin!3534	10 个月前
test_quant_mm_all_reduce_base.py	!1935 Fix some UTs on 910B Merge pull request !1935 from 张向龙3/fix_ut_910b	1 年前
test_quant_mm_all_reduce_base_bf16.py	!1988 Fix some UTs on atlas_a2 part 3 Merge pull request !1988 from 张向龙3/fix_910b_ut_3	1 年前
test_quant_mm_all_reduce_base_error_dtype.py	matmul allreduce a5 异常分支拦截 Co-authored-by: xiongyaowu<xiongyaowu@huawei.com> # message auto-generated for no-merge-commit merge: !3726 merge master into master matmul allreduce a5 异常分支拦截 Created-by: xyaowu Commit-by: xiongyaowu 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!3726	9 个月前
test_quant_mm_reduce_scatter.py	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	10 个月前
test_repeat_interleave_backward.py	!1368 RepeatInterleaveBackward 适配上库 Merge pull request !1368 from 叶和明/master	2 年前
test_rfft.py	!1429 Rfft plugin Merge pull request !1429 from chizhovkirill/RFFT	2 年前
test_save_npugraph_tensor.py	[fix] fix save_npugraph_tensor core dump when multiple replay Co-authored-by: Chukun Wang<wangchukun@huawei.com> # message auto-generated for no-merge-commit merge: !4032 merge master into master [fix] fix save_npugraph_tensor core dump when multiple replay Created-by: kita-ikuyo Commit-by: Chukun Wang Merged-by: ascend-robot Description: <!-- Thanks for sending a pull request! --> What type of PR is this? > /kind bug What does this PR do / why do we need it: fix save_npugraph_tensor core dump when multiple replay Special notes for your reviewers: 在回调函数中释放指针会触发tensor的析构，导致多次replay的时候内存地址失效导致core dump。经评估无需手动释放args。 aclgraph场景，args属于aclgraph对象，由aclgraph管理。 eager场景，args属于单次调的输入，用完由torch内存池管理。 See merge request: Ascend/op-plugin!4032	8 个月前
test_scatter_list.py	!1200 Fix 910B skip ut use decorator Merge pull request !1200 from 王广斌/master_cherry	2 年前
test_scatter_pa_kv_cache.py	!2970 add cache_mode attr for ScatterPaKvCache Merge pull request !2970 from 熊永龙/master	1 年前
test_stft_backward.py	!1935 Fix some UTs on 910B Merge pull request !1935 from 张向龙3/fix_ut_910b	1 年前
test_top_k_top_p_sample.py	Fix the documentation and UT scripts for the TopKtopPSample operator Co-authored-by: liubo<liubo483@huawei.com> # message auto-generated for no-merge-commit merge: merge master into master Fix the documentation and UT scripts for the TopKtopPSample operator Created-by: sptlb68 Commit-by: liubo 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 What does this PR do / why do we need it: pta脚本计算分支判断条件修改，md文档格式调整，与aclnn对齐 Special notes for your reviewers: See merge request: Ascend/op-plugin!3170	1 年前
test_weight_quant_mm_all_reduce_base.py	!1988 Fix some UTs on atlas_a2 part 3 Merge pull request !1988 from 张向龙3/fix_910b_ut_3	1 年前
test_weight_quant_mm_all_reduce_base_with_add.py	!1988 Fix some UTs on atlas_a2 part 3 Merge pull request !1988 from 张向龙3/fix_910b_ut_3	1 年前