fastai library v1

[miwojc]

is fastai library v1 available in kaggle kernels?
also how to switch between version v0.7 and v1 of the library in kernels, as the 'old' courses use v0.7 version and 'new' course uses v1.

[miwojc]

The kernels that were working before now don't work, so looks like now the fastai library was updated to v1 at kernels.
I wanted to install 0.7 version using 'Add custom package'
via pip: pip install fastai==0.7.0
produced error message: 'kaggle/python@sha256:...' is not a valid parent image name

via GitHub repo: git clone https://github.com/fastai/fastai.git
produced error message: 'kaggle/python@sha256:...' is not a valid parent image name

[miwojc]

When i enabled GPU in settings panel on the right it works again, so must be some difference when GPU off/on

[rosbo]

Hi,

Thank you for the feedback. I will let our Learn team know about this.

The fastai library was upgraded from 0.7.x to 1.0.x yesterday for CPU and this morning for GPU.

[rosbo]

If you want to downgrade to fastai 0.7, you can achieve this by:

1. Enabling Internet access in the Settings panel
2. Run !pip install fastai==0.7.0 in a code cell

[diskshima]

@rosbo
When you install fast 0.7, does the running Kaggle kernel reload fastai 0.7 automatically?

I'm upgrading Pytorch to 1.0 (fastai 1.0 relies on Pytorch 1.0) in my kernel but have to rely on the fact that the Kernel restarts for it to pick up the newly installed version.

[rosbo]

Hi,

We have released a new Docker image that reverts fastai back to 0.7 and keep pytorch 0.4.1.

Please, let us know if you are still experiencing issues with fastai in Kaggle Kernels

[pradgol]

Hi,
fastai is not working with pytorch 0.4.1. How can I use pytorch 0.3.1 or 1? please advise. thanks!

[rosbo]

@pradgol fastai 0.7 should work in Kaggle Kernels. Are you trying to install a different version of fastai?

[pradgol]

I am using fastai 0.7 (the default one) but it breaks when I try to fit a model. It needs pytorch 0.3.1. So I am trying to do the following but the kernel keeps dying:

1. !pip install torch==0.3.1
2. ctrl + shift+ p to restart kernel
3. import torch