Fantastic Things has Fantastic Errors

[SaraCoder777]

Hi, Your framework was the exact thing I needed, But I cant running it on Colab. Many Errors, Too many library and package mismatches and so on...
If there is Running Colab Version which could be run correctly and been able to edit Mistral, That would be nice.
But a question?
Why we couldn't use Kaggle's TPU with above 300GB of ram and much faster?
Is there any solution to run it on Kaggle at all?

Best Wishes and thanks again

[SaraCoder777]

Thanks for labeling, As a reminder I should add that The kaggle has a Direct Model Loading Section, means we can load the model, Fix and Edit it, And Save it into our working directory, and Re upload it for further use on the kaggle itself.
Thanks again

[littlefive5]

It seems that the omegaconf is incompatible with Colab, but you can click cancel when they push the warning.
https://colab.research.google.com/drive/1lBmYw7AHkwUXAUGajPer9ZPK_bAoD0GA#scrollTo=WZvRsequGCTW
I have tested the GPT-2 model, it seems that Mistral is too large and colab does not have enough RAM.

As to the Kaggle's TPU, this would take some time as we haven't tested our code on TPU environment.
Thanks.

[zxlzr]

hi buddy, have you solved your issue yet?

[SaraCoder777]

Hi, Actually not. Still trying to find a memory efficient way to using it on Kaggles TPU. Do you have some good news?

[SaraCoder777]

hi Guys, Any update?

[littlefive5]

Not much, It's also the my first time to try TPU.
Give me some time.

[SaraCoder777]

I have some findings and need to discuss them with you, I searched and couldn't find any contact. here is my Email
[email protected]
it's important.
Thanks

[littlefive5]

Sorry for being late, you can find me via [email protected]