Fast.ai Deep Learning class notes
I started taking the fast.ai deeplearning 2018 and I want to document my path.
In the videos, Jeremy recommends starting immediately by using paperspace.com but in my case I have an Ubuntu 18.0.4 installation with a Mobile NVIDIA processor, so these are the steps that I took to get it to run:
Since I was using ensorflow GPU (setup following these instructions: https://medium.com/codezillas/step-by-step-guide-to-install-tensorflow-gpu-on-ubuntu-18-04-lts-6feceb0df5c0 )
git clone email@example.com:andresesfm/fastai.git
I also had Conda installed so I didn’t just want to run the provided script. Instead, all I need was:
conda env create -f environment.yml
Ran into a bit of a hurdle,
“Found GPU0 Quadro M1000M which is of cuda capability 5.0.
PyTorch no longer supports this GPU because it is too old.”
Trying to solve folowing this:
After a long search i found the solution The problem is that if the GPU is old the pytorch version before 0.4.0 doesn't…forums.fast.ai
OK I followed the instructions to build Pytorch from the source. However, it didn’t work. I tried both version 3.0.0 and 3.1.0
Ultimately what worked was this:
install pytorch cuda91 -c pytorch
Kasianenko: cd to fastai directory in conda and writing conda env update installs pytorch 4 and this error is gone! I…forums.fast.ai
That finally returned torch.cuda.is_available() True
Now the bad news:
RuntimeError: CUDA error: out of memory
That’s as fast as I can take it on My Lenovo P50.
Note, I also attempted to run with CPU-only and it was projecting to take 43+ hours :(