Star303
157
代码
代码
Issues13
Pull Requests11
Actions
讨论
Wiki
分析
Star303
157
cann-robotcann-robot修复linear nz aclnnaddmmweightnz 场景 shape 转换
a7aeb4d9创建于 6月11日历史提交
文件最后提交记录最后更新时间
activation.csv
【A5】【ATB】支持gelu算子接入ATB框架 Co-authored-by: zouyanlong<zouyanlong@h-partners.com> # message auto-generated for no-merge-commit merge: !1994 merge opena5gelu into master 【A5】【ATB】支持gelu算子接入ATB框架 Created-by: ninihihi Commit-by: zouyanlong Merged-by: cann-robot Description: ## 描述 【A5】【ATB】支持gelu算子接入ATB框架 ## checklist - [x] 是否通过本地IDE对代码进行静态检查 - [x] 是否通过本地IDE对代码进行格式化处理 - [x] 是否进行空指针校验 - [x] 是否进行返回值校验 - [x] 是否正确释放new/malloc申请的内存 - [x] 是否充分考虑接口的异常场景 - [x] 是否正确记录错误日志 See merge request: cann/ascend-transformer-boost!1994
2 个月前
all_gather.csv
【A5】【ATB】ATB支持AllGather AllReduce ReduceScatter算子 Co-authored-by: zouyanlong<zouyanlong@h-partners.com> # message auto-generated for no-merge-commit merge: !2091 merge a5_only_hccl into master 【A5】【ATB】ATB支持AllGather AllReduce ReduceScatter算子 Created-by: ninihihi Commit-by: zouyanlong Merged-by: cann-robot Description: ## 描述 AllGather AllReduce ReduceScatter 添加A5校验 ## checklist - [x] 是否通过本地IDE对代码进行静态检查 - [x] 是否通过本地IDE对代码进行格式化处理 - [x] 是否进行空指针校验 - [x] 是否进行返回值校验 - [x] 是否正确释放new/malloc申请的内存 - [x] 是否充分考虑接口的异常场景 - [x] 是否正确记录错误日志 See merge request: cann/ascend-transformer-boost!2091
1 个月前
all_gatherv.csv
allgatherv 和 reducescatterv 添加shape校验 Co-authored-by: zouyanlong<zouyanlong@h-partners.com>
5 个月前
all_reduce.csv
【A5】【ATB】ATB支持AllGather AllReduce ReduceScatter算子 Co-authored-by: zouyanlong<zouyanlong@h-partners.com> # message auto-generated for no-merge-commit merge: !2091 merge a5_only_hccl into master 【A5】【ATB】ATB支持AllGather AllReduce ReduceScatter算子 Created-by: ninihihi Commit-by: zouyanlong Merged-by: cann-robot Description: ## 描述 AllGather AllReduce ReduceScatter 添加A5校验 ## checklist - [x] 是否通过本地IDE对代码进行静态检查 - [x] 是否通过本地IDE对代码进行格式化处理 - [x] 是否进行空指针校验 - [x] 是否进行返回值校验 - [x] 是否正确释放new/malloc申请的内存 - [x] 是否充分考虑接口的异常场景 - [x] 是否正确记录错误日志 See merge request: cann/ascend-transformer-boost!2091
1 个月前
all_to_all.csv
CI时长优化 Co-authored-by: dovahkiiin<haorunzhe@h-partners.com>
7 个月前
all_to_allv.csv
RC1 Match-id-172ae2efabeb243c83baedc7edb69d91ab60a925
1 年前
all_to_allvv2.csv
CI时长优化 Co-authored-by: dovahkiiin<haorunzhe@h-partners.com>
7 个月前
as_strided.csv
CI时长优化 Co-authored-by: dovahkiiin<haorunzhe@h-partners.com>
7 个月前
block_copy.csv
add csv test
1 年前
broadcast.csv
RC1 Match-id-172ae2efabeb243c83baedc7edb69d91ab60a925
1 年前
cohere_layernorm.csv
RC1 Match-id-172ae2efabeb243c83baedc7edb69d91ab60a925
1 年前
concat.csv
950芯片名称整改 Co-authored-by: gyq<geyunqi@huawei.com>
5 个月前
cumsum.csv
RC1 Match-id-172ae2efabeb243c83baedc7edb69d91ab60a925
1 年前
dynamic_ntk.csv
fix: dynamicNTK datagen for ntoken=1
11 个月前
elewise.csv
修复：A5 elewise_quant_per_channel 缺少对空offsetTensor的支持 Co-authored-by: zouyanlong<zouyanlong@h-partners.com>
4 个月前
fast_soft_max.csv
RC1 Match-id-172ae2efabeb243c83baedc7edb69d91ab60a925
1 年前
fast_soft_max_grad.csv
RC1 Match-id-172ae2efabeb243c83baedc7edb69d91ab60a925
1 年前
faupdate.csv
faupdate enlarge sp's upper bound
1 年前
fill.csv
950芯片名称整改 Co-authored-by: gyq<geyunqi@huawei.com>
5 个月前
fused_add_topk_div.csv
RC1 Match-id-172ae2efabeb243c83baedc7edb69d91ab60a925
1 年前
gather.csv
950芯片名称整改 Co-authored-by: gyq<geyunqi@huawei.com>
5 个月前
gather_pre_rms_norm.csv
RC1 Match-id-172ae2efabeb243c83baedc7edb69d91ab60a925
1 年前
gating.csv
CI时长优化 Co-authored-by: dovahkiiin<haorunzhe@h-partners.com>
7 个月前
gen_attention_mask.csv
CI时长优化 Co-authored-by: dovahkiiin<haorunzhe@h-partners.com>
7 个月前
gmm_deq_swiglu_quant_gmm_deq.csv
gmm 反例
1 年前
group_topk.csv
CI时长优化 Co-authored-by: dovahkiiin<haorunzhe@h-partners.com>
7 个月前
grouped_matmul_inplace_add.csv
RC1 Match-id-172ae2efabeb243c83baedc7edb69d91ab60a925
1 年前
grouped_matmul_with_routing.csv
RC1 Match-id-172ae2efabeb243c83baedc7edb69d91ab60a925
1 年前
index_add.csv
RC1 Match-id-172ae2efabeb243c83baedc7edb69d91ab60a925
1 年前
kv_cache.csv
CI时长优化 Co-authored-by: dovahkiiin<haorunzhe@h-partners.com>
7 个月前
laser_attention.csv
CI时长优化 Co-authored-by: dovahkiiin<haorunzhe@h-partners.com>
7 个月前
laser_attention_grad.csv
RC1 Match-id-172ae2efabeb243c83baedc7edb69d91ab60a925
1 年前
layer_norm.csv
[feat] support LayerNorm in Ascend950 Co-authored-by: ivanshan05075<shanzidan@h-partners.com> # message auto-generated for no-merge-commit merge: !2090 merge br_personal/ivan/layer_norm into master [feat] support LayerNorm in Ascend950 Created-by: ivanshan05075 Commit-by: ivanshan05075 Merged-by: cann-robot Description: ## 描述 <!--在这里详细描述你的改动，包括改动的原因和所采取的方法。--> ## checklist <!-- [x] 表示选中 --> - [x] 是否通过本地IDE对代码进行静态检查 - [x] 是否通过本地IDE对代码进行格式化处理 - [x] 是否进行空指针校验 - [x] 是否进行返回值校验 - [x] 是否正确释放new/malloc申请的内存 - [x] 是否充分考虑接口的异常场景 - [x] 是否正确记录错误日志 See merge request: cann/ascend-transformer-boost!2090
1 个月前
layer_norm_with_stride.csv
RC1 Match-id-172ae2efabeb243c83baedc7edb69d91ab60a925
1 年前
linear.csv
修复linear nz aclnnaddmmweightnz 场景 shape 转换 Co-authored-by: zouyanlong<zouyanlong@h-partners.com> # message auto-generated for no-merge-commit merge: !2119 merge nz_shape_align into master 修复linear nz aclnnaddmmweightnz 场景 shape 转换 Created-by: ninihihi Commit-by: zouyanlong Merged-by: cann-robot Description: ## 描述 修复addmmweightnz 场景 storageShape 转换结果不兼容的问题 fix(linear): use CANN standard NZ storageShape for WeightNz APIs ## checklist <!-- [x] 表示选中 --> - [x] 是否通过本地IDE对代码进行静态检查 - [x] 是否通过本地IDE对代码进行格式化处理 - [x] 是否进行空指针校验 - [x] 是否进行返回值校验 - [x] 是否正确释放new/malloc申请的内存 - [x] 是否充分考虑接口的异常场景 - [x] 是否正确记录错误日志 See merge request: cann/ascend-transformer-boost!2119
1 个月前
linear_parallel.csv
解决【LinearParallelOperation】【手工发现】执行LinearParallelOperation下边界值用例，报EZ9999 Co-authored-by: guangdongzhu<guanxuewei2@huawei.com>
7 个月前
linear_sparse.csv
RC1 Match-id-172ae2efabeb243c83baedc7edb69d91ab60a925
1 年前
mla_preprocess.csv
[task]add mlapo negative csv examples
10 个月前
mm_deq_swiglu_quant_mm_deq.csv
mm 反例
1 年前
multi_latent_attention.csv
feat: CheckIniMatch enhance log，added inTensors dType/format Co-authored-by: ivanshan05075<shanzidan@h-partners.com> # message auto-generated for no-merge-commit merge: !2075 merge br_personal/ivan/log_mla into master feat: CheckIniMatch 报错信息增强，打印实际入参 dtype/format Created-by: ivanshan05075 Commit-by: ivanshan05075 Merged-by: cann-robot Description: ## 描述 <!--在这里详细描述你的改动，包括改动的原因和所采取的方法。--> feat: CheckIniMatch 报错信息增强，打印实际入参 dtype/format [2026-04-30 02:41:20.858873] [error] [830216] [operation_base.cpp:408] MultiLatentAttentionOperation_0 CheckIniMatch Failed! Actual Inputs: inTensor0(int8, nd), inTensor1(float16, nd), inTensor2(int8, fractal_nz), inTensor3(float16, fractal_nz), inTensor4(int32, nd), inTensor5(int32, nd), inTensor6(float, nd), inTensor7(float, nd), outTensor0(float16, nd), outTensor1(int32, nd) Supported Combs: [comb0]:inTensor0(int8,nd), inTensor1(float16,nd), inTensor2(int8,fractal_nz), inTensor3(float16,fractal_nz), inTensor4(int32,nd), inTensor5(int32,nd), inTensor6(float,nd), inTensor7(float,nd),outTensor0(float16,nd), outTensor1(float,nd) [comb1]:inTensor0(int8,nd), inTensor1(bf16,nd), inTensor2(int8,fractal_nz), inTensor3(bf16,fractal_nz), inTensor4(int32,nd), inTensor5(int32,nd), inTensor6(float,nd), inTensor7(float,nd),outTensor0(bf16,nd), outTensor1(float,nd) [2026-04-30 02:41:20.858894] [error] [830216] [operation_base.cpp:648] MultiLatentAttentionOperation_0 invalid param, setup check fail, error code: 13 [2026-04-30 02:41:20.858907] [error] [830216] [operation_torch.cpp:225] MultiLatentAttentionOperation setup fail, not call execute, error code: 13 ## checklist <!-- [x] 表示选中 --> - [x] 是否通过本地IDE对代码进行静态检查 - [x] 是否通过本地IDE对代码进行格式化处理 - [x] 是否进行空指针校验 - [x] 是否进行返回值校验 - [x] 是否正确释放new/malloc申请的内存 - [x] 是否充分考虑接口的异常场景 - [x] 是否正确记录错误日志 See merge request: cann/ascend-transformer-boost!2075
1 个月前
multinomial.csv
add soc version in csv and cleancode
11 个月前
nonzero.csv
fix nonzero InferShapeImpl bug Co-authored-by: caixilong<caixilong2@h-partners.com>
7 个月前
norm_rope_reshape.csv
secure issue fix
10 个月前
onehot.csv
[fix] fix onehot testcase Co-authored-by: Lord_of_Ironhill<suiweiyi@huawei.com>
7 个月前
pad.csv
fix check of pad/unpad Co-authored-by: wy519<wangyang519@huawei.com>
7 个月前
pad_with_hidden_state.csv
RC1 Match-id-172ae2efabeb243c83baedc7edb69d91ab60a925
1 年前
paged_attention.csv
task: switch pa v3 to v5 Co-authored-by: qiuqianjin233<qiuqianjin@huawei.com> # message auto-generated for no-merge-commit merge: !2069 merge br_personal/ivan/pav3_v5 into master task: switch pa v3 to v5 Created-by: ivanshan05075 Commit-by: qiuqianjin233 Merged-by: cann-robot Description: ## 描述 <!--在这里详细描述你的改动，包括改动的原因和所采取的方法。--> Paged Attention 950 切换aclnnFIA从v3到v5 ## checklist <!-- [x] 表示选中 --> - [x] 是否通过本地IDE对代码进行静态检查 - [x] 是否通过本地IDE对代码进行格式化处理 - [x] 是否进行空指针校验 - [x] 是否进行返回值校验 - [x] 是否正确释放new/malloc申请的内存 - [x] 是否充分考虑接口的异常场景 - [x] 是否正确记录错误日志 See merge request: cann/ascend-transformer-boost!2069
2 个月前
paged_cache_load.csv
bugfix(paged_cache_load): modify limitation of token max size Co-authored-by: HallCin<qinhao61@h-partners.com>
8 个月前
razor_fusion_attention.csv
修复RazorFusionAttention的headSize校验bug Signed-off-by: songjianquan <songjianquan1@huawei.com>
1 年前
reduce.csv
quant_per_channel 和 reduce A5修改 Co-authored-by: zouyanlong<zouyanlong@h-partners.com>
4 个月前
reduce_scatter.csv
【A5】【ATB】ATB支持AllGather AllReduce ReduceScatter算子 Co-authored-by: zouyanlong<zouyanlong@h-partners.com> # message auto-generated for no-merge-commit merge: !2091 merge a5_only_hccl into master 【A5】【ATB】ATB支持AllGather AllReduce ReduceScatter算子 Created-by: ninihihi Commit-by: zouyanlong Merged-by: cann-robot Description: ## 描述 AllGather AllReduce ReduceScatter 添加A5校验 ## checklist - [x] 是否通过本地IDE对代码进行静态检查 - [x] 是否通过本地IDE对代码进行格式化处理 - [x] 是否进行空指针校验 - [x] 是否进行返回值校验 - [x] 是否正确释放new/malloc申请的内存 - [x] 是否充分考虑接口的异常场景 - [x] 是否正确记录错误日志 See merge request: cann/ascend-transformer-boost!2091
1 个月前
reduce_scatterv.csv
allgatherv 和 reducescatterv 添加shape校验 Co-authored-by: zouyanlong<zouyanlong@h-partners.com>
5 个月前
relay_attention.csv
RC1 Match-id-172ae2efabeb243c83baedc7edb69d91ab60a925
1 年前
repeat.csv
RC1 Match-id-172ae2efabeb243c83baedc7edb69d91ab60a925
1 年前
reshape_and_cache.csv
950 reshape and cache siso Co-authored-by: sakya_gyq<geyunqi@huawei.com> # message auto-generated for no-merge-commit merge: !2065 merge reshapeandcache into master 950 reshape and cache siso Created-by: sakya_gyq Commit-by: sakya_gyq Merged-by: cann-robot Description: ## 描述 <!--在这里详细描述你的改动，包括改动的原因和所采取的方法。--> 950接入reshapeandcache单入单出功能 ## checklist <!-- [x] 表示选中 --> - [x] 是否通过本地IDE对代码进行静态检查 - [x] 是否通过本地IDE对代码进行格式化处理 - [x] 是否进行空指针校验 - [x] 是否进行返回值校验 - [x] 是否正确释放new/malloc申请的内存 - [x] 是否充分考虑接口的异常场景 - [x] 是否正确记录错误日志 See merge request: cann/ascend-transformer-boost!2065
2 个月前
reshape_and_cache_omni.csv
RC1 Match-id-172ae2efabeb243c83baedc7edb69d91ab60a925
1 年前
ring_mla.csv
enhance ringmla error message, add csv test
1 年前
rms_norm.csv
remove eb in csv test Co-authored-by: ivanshan05075<shanzidan@h-partners.com> # message auto-generated for no-merge-commit merge: !2074 merge br_personal/ivan/remove_eb_csv into master task: remove eb precision in csv Created-by: ivanshan05075 Commit-by: ivanshan05075 Merged-by: cann-robot Description: ## 描述 <!--在这里详细描述你的改动，包括改动的原因和所采取的方法。--> csv 测试框架移除eb 用例标准 ## checklist <!-- [x] 表示选中 --> - [x] 是否通过本地IDE对代码进行静态检查 - [x] 是否通过本地IDE对代码进行格式化处理 - [x] 是否进行空指针校验 - [x] 是否进行返回值校验 - [x] 是否正确释放new/malloc申请的内存 - [x] 是否充分考虑接口的异常场景 - [x] 是否正确记录错误日志 See merge request: cann/ascend-transformer-boost!2074
2 个月前
rms_norm_backward.csv
RC1 Match-id-172ae2efabeb243c83baedc7edb69d91ab60a925
1 年前
rms_norm_with_stride.csv
RC1 Match-id-172ae2efabeb243c83baedc7edb69d91ab60a925
1 年前
rope.csv
950 rope gen, data_gen remove padding Co-authored-by: ivanshan05075<shanzidan@h-partners.com>
3 个月前
rope_grad.csv
fix(rope_grad): add check for hiddenSize Co-authored-by: HallCin<qinhao61@h-partners.com>
7 个月前
rope_q_concat.csv
RC1 Match-id-172ae2efabeb243c83baedc7edb69d91ab60a925
1 年前
scatter_elements_v2.csv
CI时长优化 Co-authored-by: dovahkiiin<haorunzhe@h-partners.com>
7 个月前
self_attention.csv
Fix all nine 310P runtime errors from the initial GQA test file: - Move KV cache and mask tensors to NPU before npu_format_cast - Convert numpy seqLen tensors via torch.from_numpy before .npu() - Use decoder BNSD data generator for decoder GQA BNSD case - Bind explicit float golden for PA encoder case - Replace unstable SelfAttentionTestDataGenerator with AttentionDataGenerator for mask-free GQA 12/1 Integrate compare_cv dual-benchmark precision alongside single-benchmark compare_output_data, aligned with test_self_attention_combine.py. Rename test_self_attention_gqa_310p.py to test_self_attention_nz.py and update the test class to TestSelfAttentionNz. Co-authored-by: Cursor <cursoragent@cursor.com> Co-authored-by: ivanshan05075<shanzidan@h-partners.com> # message auto-generated for no-merge-commit merge: !2107 merge br_personal/ivan/fa_310p_gqa2 into master test(self-attention): add 310P FA GQA NZ cases Created-by: ivanshan05075 Commit-by: ivanshan05075 Merged-by: cann-robot Description: ## 描述 <!--在这里详细描述你的改动，包括改动的原因和所采取的方法。--> SelfAttention 310p上新增GQA的用例 GQA: headNum > kvHeadNum，且headNum % kvHeadNum == 0 ## checklist <!-- [x] 表示选中 --> - [x] 是否通过本地IDE对代码进行静态检查 - [x] 是否通过本地IDE对代码进行格式化处理 - [x] 是否进行空指针校验 - [x] 是否进行返回值校验 - [x] 是否正确释放new/malloc申请的内存 - [x] 是否充分考虑接口的异常场景 - [x] 是否正确记录错误日志 ![image.png](https://raw.gitcode.com/user-images/assets/7623009/eedcbd77-3c54-4bc0-99ee-d69a5165d79e/image.png 'image.png') See merge request: cann/ascend-transformer-boost!2107
1 个月前
set_value.csv
CI时长优化 Co-authored-by: dovahkiiin<haorunzhe@h-partners.com>
7 个月前
slice.csv
修复 A5 slice 不支持uint32输入 Co-authored-by: zouyanlong<zouyanlong@h-partners.com>
3 个月前
softmax.csv
950芯片名称整改 Co-authored-by: gyq<geyunqi@huawei.com>
5 个月前
sort.csv
950芯片名称整改 Co-authored-by: gyq<geyunqi@huawei.com>
5 个月前
split.csv
add split_with_size model case Co-authored-by: sakya_gyq<geyunqi@huawei.com>
4 个月前
strided_batch_matmul.csv
RC1 Match-id-172ae2efabeb243c83baedc7edb69d91ab60a925
1 年前
swiglu_quant.csv
feat(swiglu_quant): Ascend950 ACLNN SwiGluQuantV2 with Ops-aligned scale output ## Summary Migrate SwigluQuant on Ascend 950 to CANN aclnnSwiGluQuantV2 while keeping Ascend 910B on the existing OpsRunner (AscendC) path. Unify the second output (scale) semantics with 910B via aclnnInplaceReciprocal after the ACLNN kernel. Consolidate CSV coverage into a single swiglu_quant.csv with SOC-specific cases. ## Motivation - Reuse CANN-maintained SwiGluQuantV2 on 950 instead of maintaining a separate kernel path for that platform. - Preserve upstream API compatibility: SwigluQuantParam and tensor I/O unchanged. - Keep 910B behavior and golden stable on OpsRunner (UB limits, scale = dequant). - Expose ACLNN scale as dequant scale (y_max/127) on 950 to match 910B consumers. Co-authored-by: ivanshan05075<shanzidan@h-partners.com> # message auto-generated for no-merge-commit merge: !2097 merge br_personal/ivan/switch_swiglu_quant into master feat(swiglu_quant): switch swigluquant to ACLNN aclnnSwiGluQuantV2 on 950 Created-by: ivanshan05075 Commit-by: ivanshan05075 Merged-by: cann-robot Description: ## 描述 <!--在这里详细描述你的改动，包括改动的原因和所采取的方法。--> SwigluQuant 950 芯片切换 aclnnSwiGluQuantV2 + scale 输出 拼接 aclnnInplaceReciprocal保证输出一致 ## checklist <!-- [x] 表示选中 --> - [x] 是否通过本地IDE对代码进行静态检查 - [x] 是否通过本地IDE对代码进行格式化处理 - [x] 是否进行空指针校验 - [x] 是否进行返回值校验 - [x] 是否正确释放new/malloc申请的内存 - [x] 是否充分考虑接口的异常场景 - [x] 是否正确记录错误日志 See merge request: cann/ascend-transformer-boost!2097
1 个月前
testcase_example.csv
CI时长优化 Co-authored-by: dovahkiiin<haorunzhe@h-partners.com>
7 个月前
topk_topp_sampling.csv
fix topktopp golden Co-authored-by: chenwei<chenwei542@huawei.com>
3 个月前
transdata.csv
RC1 Match-id-172ae2efabeb243c83baedc7edb69d91ab60a925
1 年前
transpose.csv
950芯片名称整改 Co-authored-by: gyq<geyunqi@huawei.com>
5 个月前
unpad.csv
fix check of pad/unpad Co-authored-by: wy519<wangyang519@huawei.com>
7 个月前
unpad_with_hidden_state.csv
RC1 Match-id-172ae2efabeb243c83baedc7edb69d91ab60a925
1 年前
where.csv
RC1 Match-id-172ae2efabeb243c83baedc7edb69d91ab60a925
1 年前