I’m looking for ideas or next steps:
I trained a fastai text classifier on a GPU machine in the cloud (Paperspace). I can use the “exported” trained model to make predictions on my Mac, after carefully ensuring that all library versions of fastai, fastcore, etc. are exactly the same on both machines.
When I try to run the trained model on my Windows 10 machine, the “load_learner” call crashes python - no stack trace, no console errors. The crash happens when running in a Jupyter notebook or directly in the python REPL.
I was not able to make my python environments exactly the same between the Mac and Windows machines. Pip couldn’t find a matching torch library, so I had to install torch based on recommendations from the torch project website.
Mac info: Catalina v10.15.5; 2.2 GHz Quad-Core Intel Core i7; 16 GB RAM; Python 3.8.5
Windows info: Win 10 Enterprise v1909; 1.7GHz Intel Core i5; 16 GB RAM; Python 3.8.5
A diff between the Mac libraries and the Windows libraries I ended up having to install is shown below.
Thanks.
— requirements.mac.txt 2020-12-04 16:03:27.000000000 -0500
+++ requirements.new.txt 2020-12-05 08:11:06.000000000 -0500
@@ -9,6 +9,7 @@
certifi==2020.11.8
cffi==1.14.3
chardet==3.0.4
+colorama==0.4.4
cycler==0.10.0
cymem==2.0.4
dataclasses==0.6
@@ -63,6 +64,8 @@
pyrsistent==0.17.3
python-dateutil==2.8.1
pytz==2020.4
+pywin32==300
+pywinpty==0.5.7
PyYAML==5.3.1
pyzmq==20.0.0
requests==2.25.0
@@ -76,8 +79,9 @@
testpath==0.4.4
thinc==7.4.1
threadpoolctl==2.1.0
-torch==1.7.0
-torchvision==0.8.1
+torch==1.7.0+cpu
+torchaudio==0.7.0
+torchvision==0.8.1+cpu
tornado==6.1
tqdm==4.53.0
traitlets==5.0.5