PyTorch 2.0 not working on Windows

[Jerry-Master]

🐛 Describe the bug

When I try installing the nightly build with the following command:

pip3 install numpy --pre torch[dynamo] --force-reinstall --extra-index-url https://download.pytorch.org/whl/nightly/cu117

It gives the following warning

WARNING: torch 2.0.0.dev20221213+cu117 does not provide the extra 'dynamo'

And when I try to check the installation with

git clone https://github.com/pytorch/pytorch
cd tools/dynamo
python verify_dynamo.py

This results in the following error

Traceback (most recent call last):
  File ".\verify_dynamo.py", line 167, in <module>
    main()
  File ".\verify_dynamo.py", line 155, in main
    cuda_ver = check_cuda()
  File ".\verify_dynamo.py", line 64, in check_cuda
    cuda_ver = get_cuda_version()
  File ".\verify_dynamo.py", line 39, in get_cuda_version
    raise VerifyDynamoError(cpp_extension.CUDA_NOT_FOUND_MESSAGE)
__main__.VerifyDynamoError:
CUDA was not found on the system, please set the CUDA_HOME or the CUDA_PATH
environment variable or add NVCC to your system PATH. The extension compilation will fail.

The torch installation is working but dynamo seems not to work. When I run the benchmarks from https://gist.github.com/Chillee/f86675147366a7a0c6e244eaa78660f7 I get the following

C:\Users\user\anaconda3\envs\auxiliar\lib\site-packages\torch\_dynamo\eval_frame.py:428: UserWarning: Windows is not currently supported, torch._dynamo.optimize() will do nothing
  warnings.warn(
eager: 1386.2895965576172us
PT 2.0: 1406.266689300537us

Which again shows that dynamo is not working. If I try installing torchdynamo apart, it gives this error

Installing collected packages: torchdynamo
  Running setup.py install for torchdynamo ... error
  error: subprocess-exited-with-error

  × Running setup.py install for torchdynamo did not run successfully.
  │ exit code: 1
  ╰─> [108 lines of output]
      running install
      C:\Users\user\anaconda3\envs\auxiliar\lib\site-packages\setuptools\command\install.py:34: SetuptoolsDeprecationWarning: setup.py install is deprecated. Use build and pip and other standards-based tools.
        warnings.warn(
      running build
      running build_py
      creating build
      creating build\lib.win-amd64-cpython-38
      creating build\lib.win-amd64-cpython-38\torchdynamo
      copying torchdynamo\allowed_functions.py -> build\lib.win-amd64-cpython-38\torchdynamo
      copying torchdynamo\bytecode_analysis.py -> build\lib.win-amd64-cpython-38\torchdynamo
      copying torchdynamo\bytecode_transformation.py -> build\lib.win-amd64-cpython-38\torchdynamo
      copying torchdynamo\codegen.py -> build\lib.win-amd64-cpython-38\torchdynamo
      copying torchdynamo\config.py -> build\lib.win-amd64-cpython-38\torchdynamo
      copying torchdynamo\convert_frame.py -> build\lib.win-amd64-cpython-38\torchdynamo
      copying torchdynamo\debug_utils.py -> build\lib.win-amd64-cpython-38\torchdynamo
      copying torchdynamo\eval_frame.py -> build\lib.win-amd64-cpython-38\torchdynamo
      copying torchdynamo\exc.py -> build\lib.win-amd64-cpython-38\torchdynamo
      copying torchdynamo\guards.py -> build\lib.win-amd64-cpython-38\torchdynamo
      copying torchdynamo\logging.py -> build\lib.win-amd64-cpython-38\torchdynamo
      copying torchdynamo\mutation_guard.py -> build\lib.win-amd64-cpython-38\torchdynamo
      copying torchdynamo\output_graph.py -> build\lib.win-amd64-cpython-38\torchdynamo
      copying torchdynamo\profiler.py -> build\lib.win-amd64-cpython-38\torchdynamo
      copying torchdynamo\replay_record.py -> build\lib.win-amd64-cpython-38\torchdynamo
      copying torchdynamo\resume_execution.py -> build\lib.win-amd64-cpython-38\torchdynamo
      copying torchdynamo\side_effects.py -> build\lib.win-amd64-cpython-38\torchdynamo
      copying torchdynamo\skipfiles.py -> build\lib.win-amd64-cpython-38\torchdynamo
      copying torchdynamo\source.py -> build\lib.win-amd64-cpython-38\torchdynamo
      copying torchdynamo\symbolic_convert.py -> build\lib.win-amd64-cpython-38\torchdynamo
      copying torchdynamo\testing.py -> build\lib.win-amd64-cpython-38\torchdynamo
      copying torchdynamo\utils.py -> build\lib.win-amd64-cpython-38\torchdynamo
      copying torchdynamo\__init__.py -> build\lib.win-amd64-cpython-38\torchdynamo
      creating build\lib.win-amd64-cpython-38\torchinductor
      copying torchinductor\codecache.py -> build\lib.win-amd64-cpython-38\torchinductor
      copying torchinductor\compile_fx.py -> build\lib.win-amd64-cpython-38\torchinductor
      copying torchinductor\config.py -> build\lib.win-amd64-cpython-38\torchinductor
      copying torchinductor\debug.py -> build\lib.win-amd64-cpython-38\torchinductor
      copying torchinductor\decomposition.py -> build\lib.win-amd64-cpython-38\torchinductor
      copying torchinductor\dependencies.py -> build\lib.win-amd64-cpython-38\torchinductor
      copying torchinductor\exc.py -> build\lib.win-amd64-cpython-38\torchinductor
      copying torchinductor\graph.py -> build\lib.win-amd64-cpython-38\torchinductor
      copying torchinductor\ir.py -> build\lib.win-amd64-cpython-38\torchinductor
      copying torchinductor\lowering.py -> build\lib.win-amd64-cpython-38\torchinductor
      copying torchinductor\metrics.py -> build\lib.win-amd64-cpython-38\torchinductor
      copying torchinductor\overrides.py -> build\lib.win-amd64-cpython-38\torchinductor
      copying torchinductor\scheduler.py -> build\lib.win-amd64-cpython-38\torchinductor
      copying torchinductor\sizevars.py -> build\lib.win-amd64-cpython-38\torchinductor
      copying torchinductor\utils.py -> build\lib.win-amd64-cpython-38\torchinductor
      copying torchinductor\virtualized.py -> build\lib.win-amd64-cpython-38\torchinductor
      copying torchinductor\__init__.py -> build\lib.win-amd64-cpython-38\torchinductor
      creating build\lib.win-amd64-cpython-38\torchdynamo\optimizations
      copying torchdynamo\optimizations\analysis.py -> build\lib.win-amd64-cpython-38\torchdynamo\optimizations
      copying torchdynamo\optimizations\backends.py -> build\lib.win-amd64-cpython-38\torchdynamo\optimizations
      copying torchdynamo\optimizations\distributed.py -> build\lib.win-amd64-cpython-38\torchdynamo\optimizations
      copying torchdynamo\optimizations\inference.py -> build\lib.win-amd64-cpython-38\torchdynamo\optimizations
      copying torchdynamo\optimizations\log_args.py -> build\lib.win-amd64-cpython-38\torchdynamo\optimizations
      copying torchdynamo\optimizations\normalize.py -> build\lib.win-amd64-cpython-38\torchdynamo\optimizations
      copying torchdynamo\optimizations\subgraph.py -> build\lib.win-amd64-cpython-38\torchdynamo\optimizations
      copying torchdynamo\optimizations\training.py -> build\lib.win-amd64-cpython-38\torchdynamo\optimizations
      copying torchdynamo\optimizations\__init__.py -> build\lib.win-amd64-cpython-38\torchdynamo\optimizations
      creating build\lib.win-amd64-cpython-38\torchdynamo\variables
      copying torchdynamo\variables\base.py -> build\lib.win-amd64-cpython-38\torchdynamo\variables
      copying torchdynamo\variables\builder.py -> build\lib.win-amd64-cpython-38\torchdynamo\variables
      copying torchdynamo\variables\builtin.py -> build\lib.win-amd64-cpython-38\torchdynamo\variables
      copying torchdynamo\variables\constant.py -> build\lib.win-amd64-cpython-38\torchdynamo\variables
      copying torchdynamo\variables\dicts.py -> build\lib.win-amd64-cpython-38\torchdynamo\variables
      copying torchdynamo\variables\functions.py -> build\lib.win-amd64-cpython-38\torchdynamo\variables
      copying torchdynamo\variables\lists.py -> build\lib.win-amd64-cpython-38\torchdynamo\variables
      copying torchdynamo\variables\misc.py -> build\lib.win-amd64-cpython-38\torchdynamo\variables
      copying torchdynamo\variables\nn_module.py -> build\lib.win-amd64-cpython-38\torchdynamo\variables
      copying torchdynamo\variables\tensor.py -> build\lib.win-amd64-cpython-38\torchdynamo\variables
      copying torchdynamo\variables\torch.py -> build\lib.win-amd64-cpython-38\torchdynamo\variables
      copying torchdynamo\variables\user_defined.py -> build\lib.win-amd64-cpython-38\torchdynamo\variables
      copying torchdynamo\variables\__init__.py -> build\lib.win-amd64-cpython-38\torchdynamo\variables
      creating build\lib.win-amd64-cpython-38\torchinductor\codegen
      copying torchinductor\codegen\autotuner.py -> build\lib.win-amd64-cpython-38\torchinductor\codegen
      copying torchinductor\codegen\common.py -> build\lib.win-amd64-cpython-38\torchinductor\codegen
      copying torchinductor\codegen\cpp.py -> build\lib.win-amd64-cpython-38\torchinductor\codegen
      copying torchinductor\codegen\triton.py -> build\lib.win-amd64-cpython-38\torchinductor\codegen
      copying torchinductor\codegen\triton_template.py -> build\lib.win-amd64-cpython-38\torchinductor\codegen
      copying torchinductor\codegen\wrapper.py -> build\lib.win-amd64-cpython-38\torchinductor\codegen
      copying torchinductor\codegen\__init__.py -> build\lib.win-amd64-cpython-38\torchinductor\codegen
      creating build\lib.win-amd64-cpython-38\torchinductor\triton_ops
      copying torchinductor\triton_ops\autotune.py -> build\lib.win-amd64-cpython-38\torchinductor\triton_ops
      copying torchinductor\triton_ops\batched_matmul.py -> build\lib.win-amd64-cpython-38\torchinductor\triton_ops
      copying torchinductor\triton_ops\conv.py -> build\lib.win-amd64-cpython-38\torchinductor\triton_ops
      copying torchinductor\triton_ops\conv1x1.py -> build\lib.win-amd64-cpython-38\torchinductor\triton_ops
      copying torchinductor\triton_ops\conv_perf_model.py -> build\lib.win-amd64-cpython-38\torchinductor\triton_ops
      copying torchinductor\triton_ops\matmul.py -> build\lib.win-amd64-cpython-38\torchinductor\triton_ops
      copying torchinductor\triton_ops\mm_perf_model.py -> build\lib.win-amd64-cpython-38\torchinductor\triton_ops
      copying torchinductor\triton_ops\utils.py -> build\lib.win-amd64-cpython-38\torchinductor\triton_ops
      copying torchinductor\triton_ops\__init__.py -> build\lib.win-amd64-cpython-38\torchinductor\triton_ops
      copying torchinductor\codegen\cpp_prefix.h -> build\lib.win-amd64-cpython-38\torchinductor\codegen
      copying torchinductor\codegen\triton_conv_delta_x.j2 -> build\lib.win-amd64-cpython-38\torchinductor\codegen
      copying torchinductor\codegen\triton_conv_delta_x_hwc.j2 -> build\lib.win-amd64-cpython-38\torchinductor\codegen
      copying torchinductor\codegen\triton_mm.j2 -> build\lib.win-amd64-cpython-38\torchinductor\codegen
      running build_ext
      building 'torchdynamo._eval_frame' extension
      creating build\temp.win-amd64-cpython-38
      creating build\temp.win-amd64-cpython-38\Release
      creating build\temp.win-amd64-cpython-38\Release\torchdynamo
      "C:\Program Files (x86)\Microsoft Visual Studio 14.0\VC\BIN\x86_amd64\cl.exe" /c /nologo /O2 /W3 /GL /DNDEBUG /MD -IC:\Users\user\anaconda3\envs\auxiliar\include -IC:\Users\user\anaconda3\envs\auxiliar\Include "-IC:\Program Files (x86)\Microsoft Visual Studio 14.0\VC\INCLUDE" "-IC:\Program Files (x86)\Microsoft Visual Studio 14.0\VC\ATLMFC\INCLUDE" "-IC:\Program Files (x86)\Windows Kits\10\include\10.0.10240.0\ucrt" /Tctorchdynamo/_eval_frame.c /Fobuild\temp.win-amd64-cpython-38\Release\torchdynamo/_eval_frame.obj -Wall
      _eval_frame.c
      C:\Program Files (x86)\Windows Kits\10\include\10.0.10240.0\ucrt\corecrt_io.h(49): warning C4820: '_finddata32i64_t': '4' bytes de relleno agregados despu‚s de miembro de datos 'name'
      C:\Program Files (x86)\Windows Kits\10\include\10.0.10240.0\ucrt\corecrt_io.h(54): warning C4820: '_finddata64i32_t': '4' bytes de relleno agregados despu‚s de miembro de datos 'attrib'
      C:\Program Files (x86)\Windows Kits\10\include\10.0.10240.0\ucrt\corecrt_io.h(64): warning C4820: '__finddata64_t': '4' bytes de relleno agregados despu‚s de miembro de datos 'attrib'
      C:\Program Files (x86)\Windows Kits\10\include\10.0.10240.0\ucrt\corecrt_io.h(69): warning C4820: '__finddata64_t': '4' bytes de relleno agregados despu‚s de miembro de datos 'name'
      c:\users\user\anaconda3\envs\auxiliar\include\pyconfig.h(205): fatal error C1083: No se puede abrir el archivo incluir: 'basetsd.h': No such file or directory
      error: command 'C:\\Program Files (x86)\\Microsoft Visual Studio 14.0\\VC\\BIN\\x86_amd64\\cl.exe' failed with exit code 2
      [end of output]

  note: This error originates from a subprocess, and is likely not a problem with pip.
error: legacy-install-failure

× Encountered error while trying to install package.
╰─> torchdynamo

note: This is an issue with the package mentioned above, not pip.
hint: See above for output from the failure.

I am using Windows 10, with AMD64 processors (the wheels for python show win_amd64 at the end). I also use 2 NVIDIA RTX 3090 with CUDA 12.0.

Versions

Collecting environment information...
PyTorch version: 2.0.0.dev20221213+cu117
Is debug build: False
CUDA used to build PyTorch: 11.7
ROCM used to build PyTorch: N/A

OS: Microsoft Windows 10 Pro
GCC version: Could not collect
Clang version: Could not collect
CMake version: Could not collect
Libc version: N/A

Python version: 3.8.15 (default, Nov 24 2022, 14:38:14) [MSC v.1916 64 bit (AMD64)] (64-bit runtime)
Python platform: Windows-10-10.0.19045-SP0
Is CUDA available: True
CUDA runtime version: Could not collect
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration: GPU 0: NVIDIA GeForce RTX 3090
Nvidia driver version: 526.86
cuDNN version: Could not collect
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True

Versions of relevant libraries:
[pip3] numpy==1.24.0rc2
[pip3] torch==2.0.0.dev20221213+cu117
[conda] numpy 1.24.0rc2 pypi_0 pypi
[conda] torch 2.0.0.dev20221213+cu117 pypi_0 pypi

cc @peterjc123 @mszhanyi @skyline75489 @nbcsm @ezyang @soumith @msaroufim @wconstab @ngimel @bdhirsh

[malfet]

[dynamo] extra requirements are redundant at the moment, but on the other hand I think neither CPU nor triton backends are available for Windows at the moment

[msaroufim]

@malfet speaking of the redundant dynamo I removed it from our main getting started blogs here pytorch/pytorch.github.io#1241 - for whatever reason though the doc build did not update @svekars

[T-Atlas]

PyTorch version: 2.0.0.dev20230104+cu117
Repeat the same problem in Windows 11

[PleezDeez]

py_cpuinfo works with 11 and gets rid of the lscpu flag when trying to install deepspeed in windows via pip. There's also a hacked together version of triton 2.0.0 some russian made I was able to use to make a working whl from

[g-i-o-r-g-i-o]

Can't install it on windows

PS D:> pip3 install numpy --pre torch --force-reinstall --index-url https://download.pytorch.org/whl/nightly/cu117
Looking in indexes: https://download.pytorch.org/whl/nightly/cu117
ERROR: Could not find a version that satisfies the requirement numpy (from versions: none)
ERROR: No matching distribution found for numpy

python version 3.9 (can't upgrade it, since it is the maximum version working with the last pytorch)

[inboxedshoe]

I think for this you can install numpy separately and then repeat the command without numpy in it. If you look in the url under numpy there's no windows binaries provided.

[Jerry-Master]

After researching a bit, the real problem is that Triton, which is the main compiler in torch 2.0, has no windows support. It is under development, in previous versions it had support, now it hasn't, I'll just wait.

[SlowFeather]

When I configured everything in Windows, I received this error:

raise RuntimeError("Windows not yet supported for torch.compile")
RuntimeError: Windows not yet supported for torch.compile

I want to know if it can run on Windows 11 + troch2.0 + cuda118 🤯