在提交新问题之前，请确保您已经在社区中搜索过相关问题，并使用了社区中提供的资源/工具后，仍未找到满意的解决方式。

⚠️ 安全信息提醒：请仔细检查提供的文本内容，确保其不包含敏感数据信息，包括但不限于：

• API 令牌或密钥
• 密码或身份验证凭证
• 私有网址或接口地址
• 个人或机密数据
• ...

在分享配置信息或代码示例时，请将敏感信息脱敏处理，或使用 <TOKEN> 等占位符替代原有内容。

环境信息

例如：
- 操作系统
- 昇腾硬件信息
- CANN软件版本
- 安装的对应软件版本

toolkit_running_version=[8.3.0.2.220:8.3.RC2]
torch 2.8.0+cpu
torch_npu 2.8.0
torchaudio 2.8.0+cpu
torchvision 0.23.0+cpu

🐛 问题描述

在8.3.RC2上报以下错误, 相关版本:
toolkit_running_version=[8.3.0.2.220:8.3.RC2]
torch 2.8.0+cpu
torch_npu 2.8.0
torchaudio 2.8.0+cpu
torchvision 0.23.0+cpu

CANN 8.2.RC1没有问题

Welcome to ask questions and discuss with other members.

问题描述

✅ torch.stft with n_fft=320 - Output size: torch.Size([161, 3001]), device: npu:0, output dtype: torch.complex64
✅ torch.stft with n_fft=480 - Output size: torch.Size([241, 2001]), device: npu:0, output dtype: torch.complex64
✅ torch.stft with n_fft=640 - Output size: torch.Size([321, 1501]), device: npu:0, output dtype: torch.complex64
✅ torch.stft with n_fft=960 - Output size: torch.Size([481, 1001]), device: npu:0, output dtype: torch.complex64
✅ torch.stft with n_fft=1920 - Output size: torch.Size([961, 501]), device: npu:0, output dtype: torch.complex64
✅ torch.stft with n_fft=512 - Output size: torch.Size([257, 1876]), device: npu:0, output dtype: torch.complex64
✅ torch.stft with n_fft=1024 - Output size: torch.Size([513, 938]), device: npu:0, output dtype: torch.complex64
✅ torch.stft with n_fft=2048 - Output size: torch.Size([1025, 469]), device: npu:0, output dtype: torch.complex64
✅ torch.stft with n_fft=4096 - Output size: torch.Size([2049, 235]), device: npu:0, output dtype: torch.complex64

[W326 19:25:01.837237151 GetCANNInfo.cpp:117] Warning: Driver Version: X[.@h is invalid or not supported yet. (function operator())
✅ torch.istft with n_fft=320 - Output size: torch.Size([480000]), device: npu:0, output dtype: torch.float32
✅ torch.istft with n_fft=480 - Output size: torch.Size([480000]), device: npu:0, output dtype: torch.float32
✅ torch.istft with n_fft=640 - Output size: torch.Size([480000]), device: npu:0, output dtype: torch.float32
✅ torch.istft with n_fft=960 - Output size: torch.Size([480000]), device: npu:0, output dtype: torch.float32
❌ torch.istft with n_fft=1920 - Error: unfold_backward:third_party/op-plugin/op_plugin/ops/opapi/UnfoldBackwardLernelNpuOpApi.cpp:38 NPU function error: call aclnnUnfoldGrad failed, error code is 561002
[ERROR] 2026-03-26-19:25:02 (PID:332703, Device:0, RankID:-1) ERR00100 PTA call acl api failed.
EZ9999: Inner Error!
EZ9999[PID: 332703] 2026-03-26-19:25:02.203.151 (EZ9999): TasksOnceMaxPerCore 0 must be greater than 0.[FUNC:Tiling4Block][FILE:unfold_grad_tiling.cpp][LINE:300]
TraceBack (most recent call last):
Tiling4Block failed.[FUNC:DoTiling][FILE:unfold_grad_tiling.cpp][LINE:374]
UnfoldGrad do tiling failed, ret is -1.
Check NnopbaseExecutorDoTiling(executor) failed
Check NnopbaseExecutorTilingAndUpdateBinInfo(executor) failed
Check NnopbaseExecutorMatchCache(executor) failed
Check NnopbaseRunForWorkspace(*executor, workspaceSize) failed

✅ torch.istft with n_fft=512 - Output size: torch.Size([480000]), device: npu:0, output dtype: torch.float32
✅ torch.istft with n_fft=1024 - Output size: torch.Size([480000]), device: npu:0, output dtype: torch.float32
❌ torch.istft with n_fft=2048 - Error: unfold_backward:third_party/op-plugin/op_plugin/ops/opapi/UnfoldBackwardLernelNpuOpApi.cpp:38 NPU function error: call aclnnUnfoldGrad failed, error code is 561002
[ERROR] 2026-03-26-19:25:02 (PID:332703, Device:0, RankID:-1) ERR00100 PTA call acl api failed.
EZ9999: Inner Error!
EZ9999[PID: 332703] 2026-03-26-19:25:02.257.913 (EZ9999): TasksOnceMaxPerCore 0 must be greater than 0.[FUNC:Tiling4Block][FILE:unfold_grad_tiling.cpp][LINE:300]
TraceBack (most recent call last):
Tiling4Block failed.[FUNC:DoTiling][FILE:unfold_grad_tiling.cpp][LINE:374]
UnfoldGrad do tiling failed, ret is -1.
Check NnopbaseExecutorDoTiling(executor) failed
Check NnopbaseExecutorTilingAndUpdateBinInfo(executor) failed
Check NnopbaseExecutorMatchCache(executor) failed
Check NnopbaseRunForWorkspace(*executor, workspaceSize) failed

❌ torch.istft with n_fft=4096 - Error: unfold_backward:third_party/op-plugin/op_plugin/ops/opapi/UnfoldBackwardLernelNpuOpApi.cpp:38 NPU function error: call aclnnUnfoldGrad failed, error code is 561002
[ERROR] 2026-03-26-19:25:02 (PID:332703, Device:0, RankID:-1) ERR00100 PTA call acl api failed.
EZ9999: Inner Error!
EZ9999[PID: 332703] 2026-03-26-19:25:02.402.132 (EZ9999): TasksOnceMaxPerCore 0 must be greater than 0.[FUNC:Tiling4Block][FILE:unfold_grad_tiling.cpp][LINE:300]
TraceBack (most recent call last):
Tiling4Block failed.[FUNC:DoTiling][FILE:unfold_grad_tiling.cpp][LINE:374]
UnfoldGrad do tiling failed, ret is -1.
Check NnopbaseExecutorDoTiling(executor) failed
Check NnopbaseExecutorTilingAndUpdateBinInfo(executor) failed
Check NnopbaseExecutorMatchCache(executor) failed
Check NnopbaseRunForWorkspace(*executor, workspaceSize) failed

欢迎加入社区，感谢您对社区的贡献 🎉!