Tensor is int32 error when running fit

I’m getting some errors running fastai 1.0.58 pytorch 1.0.0 on windows as listed below

C:\ProgramData\Anaconda3\envs\fastai_v1\lib\site-packages\fastai\torch_core.py:83: UserWarning: Tensor is int32: upgrading to int64; for better performance use int64 input

when running fit on a cnn_learner. Things still work, but it is kind of annoying. Any suggestions would be appreciated.

1 Like

Did you ever figure this out? I get it too.

Windows 10, PyTorch 1.2.0

No, I just ignored the warning and moved on. :frowning:

For me, inference on Google Colab (Ubuntu 18.04) w/ CPU is about 8x faster than Windows 10 w/ CPU, even though the CPUs have similar benchmarks. I’m not sure it is the int32 tensor problem, but I noticed when doing batch inference on Windows, fastai doesn’t use all the CPUs, so it seems like a multiprocessing issue. I posted a separate thread about it.

1 Like