Hey @pierreguillou, you will need to do conda uninstall --force
this long wait between epochs seems to be related to pytorch issue on windows, documented here Unexplained long delay between batches with little/no CPU, GPU, or Disk activity
i have set num_workers=0 and now epochs start instantly and there is no long wait but the epochs take longer. so in total it took 10 minutes. which is worse than before…
1 Like
Hello @partho. Thanks for your return. Could you confirm that the following code is correct in order to uninstall pytorch, install the pytorch version 1.0.0 and keep installed fastai and torchvision ?
conda uninstall pytorch --force
conda install pytorch=1.0.0
[ EDIT ] I tested and it works great! With conda uninstall pytorch --force, no need to reinstall all libraries but just pytorch. Thanks Partho 
Hi @miwojc. I’ve updated my medium post with num_workers=0 in the databunch in order to improve the training time by epoch on windows pytorch but I wrote that others parameters could alter this as well.
If @sgugger could give us his point of view, it would be very helpul
Thanks @pierreguillou,
Your instructions solved my problem about GPU not available(described in post above ).
In my case, I downgraded pytorch version from 1.0.1 to 1.0.0 with the command below:
conda uninstall pytorch
conda install fastai pytorch=1.0.0 -c fastai -c pytorch -c conda-forge
Be aware that the command “conda uninstall pytorch” will remove fastai, pytorch and torchvision.
That’s why , in the second command above, you have to install fastai and pytorch.
Now , Fastai is running smoothly on my windows 10 laptop with 1070 GPU.
My installation is like this:
=== Software ===
python : 3.6.8
fastai : 1.0.46
fastprogress : 0.1.20
torch : 1.0.0
torch cuda : 9.0 / is available
torch cudnn : 7005 / is enabled
=== Hardware ===
torch devices : 1
- gpu0 : GeForce GTX 1070 with Max-Q Design
=== Environment ===
platform : Windows-10-10.0.17134-SP0
conda env : fastai_v1
python : C:\Users\Fernando\Anaconda3\envs\fastai_v1\python.exe
sys.path :
C:\Users\Fernando\Anaconda3\envs\fastai_v1\python36.zip
C:\Users\Fernando\Anaconda3\envs\fastai_v1\DLLs
C:\Users\Fernando\Anaconda3\envs\fastai_v1\lib
C:\Users\Fernando\Anaconda3\envs\fastai_v1
C:\Users\Fernando\Anaconda3\envs\fastai_v1\lib\site-packages
no nvidia-smi is found
1 Like
Hello @NandoBr. Thank you for your message and happy to read that you have solved your problem
An improvement: thanks to @partho, there is a simplest solution to change the pytorch version without removing then reinstalling all libraries.
conda uninstall pytorch --force
conda install pytorch=1.0.0
I’ve updated my medium post with this code.
I encountered the same problem as the person starting this thread: whatever I try: conda: command not found. I do see a lot of conda files in the directory fastai/conda.
Did you encounter any of this at all?
Thanks!
I think perhaps anaconda is not correctly added to your PATH variable?
Thanks so much for the guide, it allowed me to open the fastai Jupyter notebook after following it. Hopefully I don’t get any errors running the course! I’ll report back when I’m through it all.
1 Like
Happy 
1 Like
Hi, I am trying to install Fast AI from a windows host PC onto AWS and after the SSH phase, I try to
conda update conda
I get a error of “conda: command not found”
Did you find a solution?
I did the whole ML in 2018 and I didn’t have any problems back then…
The new conda install is not added to path by default anymore. So try to add it to your path.
Thank you for the reply but I’ve found out that I needed to
source ~/.bashrc
before
conda update conda
Hello @pierreguillou,
I’m having the same issue with ModuleNotFoundError: No module named 'torch' when I run the first cell. In the Anaconda3 Prompt I entered python -m fastai.utils.show_install and got:
C:\Users\92175\Anaconda3\python.exe: Error while finding module specification for 'fastai.utils.show_install' (ModuleNotFoundError: No module named 'fastai.utils')
However when I click on Environments in Anaconda Navigator, I see fastai as one of them, and I have all 5 packages installed (torch, pytorch, torchvision, etc). Any idea what might be wrong?
“show_install” moved.
From a repl or notebook, you can import it and run it as follows:
from fastai import test_utils; fastai.test_utils.show_install(1)
or from a shell prompt you can run
python -c "from fastai import test_utils; fastai.test_utils.show_install(1)"
Did you drop a symbolic link to fastai/fastai in the course folder?
Not doing this caused a similar problem for me.