Fastai v2 chat

Use this topic for general chat about fastai v2 dev - especially questions and answers that are fairly short, and for real-time discussion.


@sgugger @radek @ashaw @AlexisGallagher - mentioning you all so you know this is here…


Super excited to dive into fastai v2! Thanks to the whole fastai team for all of your work building this.

What would you want to see from fastai_audio to merge it into fastai v2 as an application (aka Do you have a requirement that things are built up using notebooks? Just trying to judge how many modifications we will need to make to the current code-base. For example, if it works with v2, but it hasn’t been built up using the Jupyter Notebook method of development, could it still get merged into fastai v2 or is that a requirement to be added as an application of fastai.

Looking forward to one day using without having to use a symlink! :slight_smile:

1 Like

Yup it needs to be a notebook, since we needs tests and docs - and it’s so much easier when they’re all in one place!


Sounds good, thanks!

Would it be appropriate to create another wiki for fastai v2 audio module or something for future module discussions?

Yup! .

1 Like

Is there a requirements.txt or a list of external library requirements or is that something that would be helpful for fastai_dev?

Edit: Created a PR to add environment.yml

1 Like

Yes that would be helpful.

Excited to dive into fast V2, cloned the repo and just finished setting it up and running the first two notebooks, just to check that i have most dependencies taken care of. I agree a requirements text would be great :slightly_smiling_face: @KevinB

1 Like

BTW note that the last few notebooks (starting with 9) are actually kinda the first - they’re the ones that set up that notebook infrastructure.


Is there a pattern to the rest of the numbering? It looks like text is the 30s and tabular is the 40s, anything else?

Can anybody tell me what version of typeguard is being used for current development?

The version I’m getting by default from conda is giving me errors so I am guessing it’s a version issue.

Is there an environment.yml (or requirements.txt) file we can/should use so that we’re all working with the appropriate packages and versions?

I have a question regarding best practices for code navigation. When developping in libraries written in .py files, one can use standard ide tools to navigate to function definition, find usages etc…

What do you recommend to navigate through the notebooks?

I am working on creating this. Here is what I’ve got so far (gettting errors on nb 01):

I don’t expect this to be exhaustive yet, but I’m trying to only add libraries as I get an error by not having them so it hopefully doesn’t get too bloated with extras.


Does anybody know what version of python fastaiv2 was developed with?

I think I found my issue (pytorch<0.4 requirement)

fastai_v2 appears to have been developed with python 3.7+

Curious if people have thoughts on this:

I’m not really sure how to handle that without just installing fastai, but it seems weird to have to install fastai_v1 for a script to work in fastai_v2. Is there a better way to do this without depending on v1?

1 Like

How related is fastai_v2’s development with SwiftAI? Is SwiftAI planned to mirror fastai_v2 from the get go, or is it more a case of: get Swift working, and _v2 will be the new standard to port when the time comes?

Hi Kevin!
Could I please ask how you’re able to find what needs to developed in V2? Is this through GitHub issues?

Sorry, might be a layman question, but how do you know where to start working? I really want to contribute.