microsoft / DirectML

DirectML is a high-performance, hardware-accelerated DirectX 12 library for machine learning. DirectML provides GPU acceleration for common machine learning tasks across a broad range of supported hardware and drivers, including all DirectX 12-capable GPUs from vendors such as AMD, Intel, NVIDIA, and Qualcomm.
MIT License
2.24k stars 299 forks source link

The operator 'aten::native_dropout_backward' is not currently supported on the DML backend #630

Open maikelsz opened 3 months ago

maikelsz commented 3 months ago

I uninstalled and then upgraded VSCode, Pytorch and DML.

a very simple and small model, using torch.nn.Dropout

:\Users\User\AppData\Local\Programs\Python\Python312\Lib\site-packages\torch\autograd\graph.py:744: UserWarning: The operator 'aten::native_dropout_backward' is not currently supported on the DML backend and will fall back to run on the CPU. This may have performance implications. (Triggered internally at C:__w\1\s\pytorch-directml-plugin\torch_directml\csrc\dml\dml_cpu_fallback.cpp:17.) return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass

....and now the work do fallback to the CPU...

before that I ´ve been running on the GPU everything without any problem for a couple of years.

should I go back to older versions?

AMD gpu Windows 10 VSCode 1.88.1 Pytorch 3.12.5 torch-directml-0.2.4.dev240815