torch version to run a demo code for FNet

[ra312]

Hi guys! Thanks for the awesome work! I am excited to play with the code. Here is the gist
https://gist.github.com/ra312/f3c895aba6e8954985258de10e9be52f
At the first attempt, I am encountered this exception

I believe the right version of pytorch shold help to fix this!
These are my current dependencies

If someone can advise on torch version, it would save me time and be super cool!
Many thanks, Rauan.

[dnairns-arch]

I ran into the same exact issue. I deleted the min_freq from the nlp_classification.py and then I ended up with another error.

IndexError: index out of range in self

Seems to be more than one issue with the FNet code...

[vpj]

Hi,

I am sorry, I don't know why you get this error.

Vocab accepts parameter min_freq - documentation, and you have the latest version (same as documentation).

Can you try initializing torchtext.vocab.Vocab(Counter(), min_freq=1) and see if you still get this error?

[vpj]

Ok, great I could reproduce the error on google colab. I'll see whats going on and get back to you.

[vpj]

found it, looks like its a function vocab class Vocab used to take these parameters as input in older torchtext versions. Will upgrade my setup and update the code. Thanks for opening the issue.

[vpj]

Looks like torchtext=0.10 requires torch=1.9, can you try downgrading to torchtext=0.9 for now (which uses torch=1.8)?.

I will have to test all experiments and models before moving to torch=1.9.

Thanks

[dnairns-arch]

Downgraded the versions to torch==1.8, and torchtext==0.9.

Re-added min_freq=1 back to nlp-classification

The FNet model is now training.

I was getting my error due to the same issue, and having removed the min_freq arg.

Let me know if you need anything else! Thanks for the help! (I am not the OP so you might wait to close until he confirms fix too.)