Ascend/pytorch

Issues

• 昇腾310p LLM推理报错：RuntimeError: copy_d2d:build/CMakeFiles/torch_npu.dir/compiler_depend.ts:104 NPU function error:

  #49 opened 2 months ago by Serious-H
  5

• 环境变量设置HCCL_EXEC_TIMEOUT=0不生效

  #54 opened a month ago by bombomla
  0

• AttributeError: module 'torch_npu.npu' has no attribute 'profile'

  #53 opened a month ago by vfdff
  1

• torch_npu aten::_transformer_encoder_layer_fwd, 运行时有警告出现，该模块未使用NPU而是使用CPU计算

  #51 opened 3 months ago by high426
  2

• torch.nn.ConvTranspose3d 不支持计算

  #52 opened 3 months ago by high426
  1

• 在使用Ascend进行模型训练的时候遇到错误

  #48 opened 3 months ago by LianShuaiLong
  1

• 链接时报错torch_npu::init_npu(std::string const&)

  #41 opened 5 months ago by nihao1997
  5

• replace -> repalce，无伤大雅的笔误

  #50 opened 3 months ago by cillinzhang
  0

• call aclnnInplaceMaskedFillScalar failed

  #45 opened 4 months ago by KirtoXX
  1

• 使用NPU推理Transfomers模型，AiCore占用为0

  #44 opened 4 months ago by Xu-backup
  1

• 2.1.0版本build报错

  #47 opened 4 months ago by nihao1997
  1

• Met error when distributed training

  #43 opened 5 months ago by ChrisMii
  1

• Error occurs when I run stable diffusion 2.1

  #42 opened 5 months ago by HoiM
  5

• no module named torch_npu._C

  #24 opened 8 months ago by ShiBoyu0204
  3

• Lora训练stable-diffusion报pytorch错误：AttributeError: module 'torch_npu' has no attribute 'npu_geglu'

  #23 opened 9 months ago by BruceWang1996
  3

• torch-npu 不支持算子 aten::isin.Tensor_Tensor_out

  #40 opened 5 months ago by natureLanguageQing
  1

• 创建一npu上的empty tensor

  #39 opened 5 months ago by nihao1997
  1

• 训练成功，但是推理服务报错

  #34 opened 6 months ago by apachemycat
  2

• euleros based系统上无法通过pip安装 torch-npu

  #38 opened 5 months ago by statelesshz
  1

• Huawei NPU device_map=auto doesn't split model evenly over all devices

  #31 opened 7 months ago by sunnyregion
  3

• import torch_npu时报错：AttributeError: module 'torch_npu.dynamo.torchair' has no attribute 'get_npu_backend'

  #37 opened 6 months ago by zywhehe
  1

• libtorch_npu.so: undefined symbol: _ZN5torch10ValueErrorC1EPKcz'!

  #35 opened 6 months ago by apachemycat
  1

• 使用torch-npu时比在gpu上跑推理要慢太多？

  #33 opened 6 months ago by 13102180531
  1

• pip install flash_attn 提示

  #32 opened 6 months ago by CloudWise-Lukemiao
  1

• 您好，使用torch_npu时报错

  #7 opened a year ago by HankLiu10
  10

• 有类似flash-attn的npu实现吗？F.scaled_dot_product_attention？

  #25 opened 7 months ago by janelu9
  1

• 请问op_pulgin的最终实现在哪？

  #30 opened 7 months ago by lms-mt
  1

• pytorch_npu-2.1 可以推理，训练出错，tbe.common无法加载

  #29 opened 8 months ago by sly123197811
  5

• 是否有查看显存占用相关方法的实现？(如torch.cuda.mem_get_info)

  #28 opened 8 months ago by RrankPyramid
  3

• Profile 拿到的 trace_view.json 不合法

  #27 opened 8 months ago by Yangruipis
  1

• [BUG] test/network_ops/test_apply_adam.py has nans in var_o for test_apply_adam_out_fp32

  #21 opened 9 months ago by nrmer
  1

• Convergence on SFT is too slow and the performance is bad

  #17 opened a year ago by Kunhao18
  1

• when i setup MultiScaleDeformableAttention, this problem happen.(about BiShengExtension)

  #26 opened 8 months ago by chaossshan
  1

• torch_npu打印tensor报错

  #14 opened a year ago by anaivebird
  2

• 请问torch_npu._C 的实现代码在哪个路径下

  #19 opened a year ago by shysuen
  1

• [BUG] multi-npu training error：Could not compute stream ID for XXX on device -1

  #20 opened 10 months ago by SolenoidWGT
  1

• RuntimeError: Failed to make directory fo profiling

  #22 opened 9 months ago by AlbertZhangHIT
  1

• pip failed to install torch-npu

  #18 opened a year ago by sxzhang2009
  1

• torch.Tensor.to(<int num>) failed on npu

  #16 opened a year ago by statelesshz
  1

• 不支持多线程并发执行，cuda 版本我看是可以的

  #13 opened a year ago by junior-zsy
  15

• 使用NPU+Pytorch2.1.0+SFTTrainer训练时tensor未完全加载于NPU

  #10 opened a year ago by DataLynnAI
  2

• How to get npu device and do smoke test

  #6 opened a year ago by GuWei007
  4

• 文档错误

  #11 opened a year ago by wpc2333
  1

• Trigger for CI

  #12 opened a year ago by FFFrog
  2

• AttributeError: module 'torch_npu' has no attribute '_isinstance'

  #9 opened a year ago by liyifango
  1

• Unsupported data type for HCCL process group

  #3 opened a year ago by iiacobac
  3

• 有支持inductor + triton功能吗

  #5 opened a year ago by peaceorwell
  2

• test_div.py一直卡着不动

  #1 opened a year ago by llhuii
  2

• 有支持的API手册吗

  #4 opened a year ago by GuWei007
  2

• 请问torch_npu怎么构建？

  #2 opened 2 years ago by moowcharnfu
  1