Apple M1 Pro环境下PyTorch报NotImplementedError的排查与解决咨询

设备配置 #

• Apple M1 Pro芯片
• 16GB内存
• macOS 13.0.1系统
• PyTorch版本：torch 1.13.0、torchvision 0.14.0（通过pip安装）

运行的Python代码 #

import torch 
mps_device = torch.device("mps")
x = torch.ones(2, device=mps_device)
x.to("cpu").pin_memory()

报错信息 #

Traceback (most recent call last):
  File "/Users/usr/path/to/file.py", line 4, in <module>
    x.to("cpu", non_blocking=True).pin_memory()
NotImplementedError: Could not run 'aten::_pin_memory' with arguments from the 'CUDA' backend. This could be because the operator doesn't exist for this backend, or was omitted during the selective/custom build process (if using custom build). If you are a Facebook employee using PyTorch on mobile, please visit https://fburl.com/ptmfixes for possible resolutions. 'aten::_pin_memory' is only available for these backends: [MPS, BackendSelect, Python, FuncTorchDynamicLayerBackMode, Functionalize, Named, Conjugate, Negative, ZeroTensor, ADInplaceOrView, AutogradOther, AutogradCPU, AutogradCUDA, AutogradHIP, AutogradXLA, AutogradMPS, AutogradIPU, AutogradXPU, AutogradHPU, AutogradVE, AutogradLazy, AutogradMeta, AutogradPrivateUse1, AutogradPrivateUse2, AutogradPrivateUse3, AutogradNestedTensor, Tracer, AutocastCPU, AutocastCUDA, FuncTorchBatched, FuncTorchVmapMode, Batched, VmapMode, FuncTorchGradWrapper, PythonTLSSnapshot, FuncTorchDynamicLayerFrontMode, PythonDispatcher].

MPS: registered at /Users/runner/work/pytorch/pytorch/pytorch/build/aten/src/ATen/RegisterMPS.cpp:20632 [kernel]
BackendSelect: registered at /Users/runner/work/pytorch/pytorch/pytorch/build/aten/src/ATen/RegisterBackendSelect.cpp:726 [kernel]
Python: registered at /Users/runner/work/pytorch/pytorch/pytorch/aten/src/ATen/core/PythonFallbackKernel.cpp:140 [backend fallback]
FuncTorchDynamicLayerBackMode: registered at /Users/runner/work/pytorch/pytorch/pytorch/aten/src/ATen/functorch/DynamicLayer.cpp:488 [backend fallback]
Functionalize: registered at /Users/runner/work/pytorch/pytorch/pytorch/aten/src/ATen/FunctionalizeFallbackKernel.cpp:291 [backend fallback]
Named: registered at /Users/runner/work/pytorch/pytorch/pytorch/aten/src/ATen/core/NamedRegistrations.cpp:7 [backend fallback]
Conjugate: registered at /Users/runner/work/pytorch/pytorch/pytorch/aten/src/ATen/ConjugateFallback.cpp:18 [backend fallback]
Negative: registered at /Users/runner/work/pytorch/pytorch/pytorch/aten/src/ATen/native/NegateFallback.cpp:18 [backend fallback]
ZeroTensor: registered at /Users/runner/work/pytorch/pytorch/pytorch/aten/src/ATen/ZeroTensorFallback.cpp:86 [backend fallback]
ADInplaceOrView: fallthrough registered at /Users/runner/work/pytorch/pytorch/pytorch/aten/src/ATen/core/VariableFallbackKernel.cpp:64 [backend fallback]
AutogradOther: registered at /Users/runner/work/pytorch/pytorch/pytorch/torch/csrc/autograd/generated/VariableType_0.cpp:14904 [autograd kernel]
AutogradCPU: registered at /Users/runner/work/pytorch/pytorch/pytorch/torch/csrc/autograd/generated/VariableType_0.cpp:14904 [autograd kernel]
AutogradCUDA: registered at /Users/runner/work/pytorch/pytorch/pytorch/torch/csrc/autograd/generated/VariableType_0.cpp:14904 [autograd kernel]
AutogradHIP: registered at /Users/runner/work/pytorch/pytorch/pytorch/torch/csrc/autograd/generated/VariableType_0.cpp:14904 [autograd kernel]
AutogradXLA: registered at /Users/runner/work/pytorch/pytorch/pytorch/torch/csrc/autograd/generated/VariableType_0.cpp:14904 [autograd kernel]
AutogradMPS: registered at /Users/runner/work/pytorch/pytorch/pytorch/torch/csrc/autograd/generated/VariableType_0.cpp:14904 [autograd kernel]
AutogradIPU: registered at /Users/runner/work/pytorch/pytorch/pytorch/torch/csrc/autograd/generated/VariableType_0.cpp:14904 [autograd kernel]
AutogradXPU: registered at /Users/runner/work/pytorch/pytorch/pytorch/torch/csrc/autograd/generated/VariableType_0.cpp:14904 [autograd kernel]
AutogradHPU: registered at /Users/runner/work/pytorch/pytorch/pytorch/torch/csrc/autograd/generated/VariableType_0.cpp:14904 [autograd kernel]
AutogradVE: registered at /Users/runner/work/pytorch/pytorch/pytorch/torch/csrc/autograd/generated/VariableType_0.cpp:14904 [autograd kernel]
AutogradLazy: registered at /Users/runner/work/pytorch/pytorch/pytorch/torch/csrc/autograd/generated/VariableType_0.cpp:14904 [autograd kernel]
AutogradMeta: registered at /Users/runner/work/pytorch/pytorch/pytorch/torch/csrc/autograd/generated/VariableType_0.cpp:14904 [autograd kernel]
AutogradPrivateUse1: registered at /Users/runner/work/pytorch/pytorch/pytorch/torch/csrc/autograd/generated/VariableType_0.cpp:14904 [autograd kernel]
AutogradPrivateUse2: registered at /Users/runner/work/pytorch/pytorch/pytorch/torch/csrc/autograd/generated/VariableType_0.cpp:14904 [autograd kernel]
AutogradPrivateUse3: registered at /Users/runner/work/pytorch/pytorch/pytorch/torch/csrc/autograd/generated/VariableType_0.cpp:14904 [autograd kernel]
AutogradNestedTensor: registered at /Users/runner/work/pytorch/pytorch/pytorch/torch/csrc/autograd/generated/VariableType_0.cpp:14904 [autograd kernel]
Tracer: registered at /Users/runner/work/pytorch/pytorch/pytorch/torch/csrc/autograd/generated/TraceType_0.cpp:16458 [kernel]
AutocastCPU: fallthrough registered at /Users/runner/work/pytorch/pytorch/pytorch/aten/src/ATen/autocast_mode.cpp:482 [backend fallback]
AutocastCUDA: fallthrough registered at /Users/runner/work/pytorch/pytorch/pytorch/aten/src/ATen/autocast_mode.cpp:324 [backend fallback]
FuncTorchBatched: registered at /Users/runner/work/pytorch/pytorch/pytorch/aten/src/ATen/functorch/LegacyBatchingRegistrations.cpp:743 [backend fallback]
FuncTorchVmapMode: fallthrough registered at /Users/runner/work/pytorch/pytorch/pytorch/aten/src/ATen/functorch/VmapModeRegistrations.cpp:28 [backend fallback]
Batched: registered at /Users/runner/work/pytorch/pytorch/pytorch/aten/src/ATen/BatchingRegistrations.cpp:1064 [backend fallback]
VmapMode: fallthrough registered at /Users/runner/work/pytorch/pytorch/pytorch/aten/src/ATen/VmapModeRegistrations.cpp:33 [backend fallback]
FuncTorchGradWrapper: registered at /Users/runner/work/pytorch/pytorch/pytorch/aten/src/ATen/functorch/TensorWrapper.cpp:189 [backend fallback]
PythonTLSSnapshot: registered at /Users/runner/work/pytorch/pytorch/pytorch/aten/src/ATen/core/PythonFallbackKernel.cpp:148 [backend fallback]
FuncTorchDynamicLayerFrontMode: registered at /Users/runner/work/pytorch/pytorch/pytorch/aten/src/ATen/functorch/DynamicLayer.cpp:484 [backend fallback]
PythonDispatcher: registered at /Users/runner/work/pytorch/pytorch/pytorch/aten/src/ATen/core/PythonFallbackKernel.cpp:144 [backend fallback]

用户疑问 #

1. 如何检查MPS后端是否可用？
2. 上述报错是否可以解决？

一、检查MPS后端可用性 #

运行以下代码即可快速确认：

import torch
print(torch.backends.mps.is_available())  # 输出True表示设备支持MPS
print(torch.backends.mps.is_built())      # 输出True表示当前PyTorch版本包含MPS支持

你的设备是M1 Pro+macOS13.0.1，满足MPS最低要求（Apple Silicon芯片+macOS12.0以上），且torch1.13.0已支持MPS，因此两个值应均为True。

二、解决pin_memory报错问题 #

原因分析 #

报错核心是：从MPS设备转移到CPU的张量，不支持pin_memory()操作。pin_memory()原本是为CUDA设备设计的，用于加速CPU到CUDA的数据传输，但MPS与CUDA的内存模型不同，torch1.13.0版本尚未实现MPS→CPU张量的pin_memory支持。

解决方法 #

根据使用场景，可选择以下两种方案：

1. 直接移除pin_memory()调用：
   若无需固定CPU张量内存（比如不需要后续快速传输回MPS），直接删除该方法即可：

   import torch 
   mps_device = torch.device("mps")
   x = torch.ones(2, device=mps_device)
   x = x.to("cpu")  # 移除pin_memory()
   

2. 多设备兼容处理：
   若代码需要同时兼容CUDA和MPS环境，可添加条件判断：

   import torch 
   mps_device = torch.device("mps")
   x = torch.ones(2, device=mps_device)
   x = x.to("cpu")
   if torch.cuda.is_available():
       x = x.pin_memory()
   

版本升级建议 #

若必须使用pin_memory()，可尝试升级PyTorch到2.0及以上版本，后续版本对MPS的支持更完善，已修复该兼容性问题。

内容的提问来源于stack exchange，提问作者Rgkpdx