New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add t5 adapter #182
Add t5 adapter #182
Conversation
… is_decoder is also true.
…orConditionalGeneration.
@AmirAktify thanks for working on this! My research team was looking to use adapters with T5 and we found this branch you've been working on. Do you think this is in a usable stage? Thanks! |
Hi Deniz. I was running some tests yesterday with sequence classification with T5ForConditionalGeneration, and I ran into some minor errors with the base model. Lemme try and debug those today, but I am hopefully very close. |
@AmirAktify That's great to hear. Thanks for letting me know so quickly. Good luck! |
Done, thank you. |
Thanks for finishing this PR btw! I hope it wasn't quite as bad as just doing it from scratch. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Looks nice overall :)
Besides the line-specific comments, two general points from the implementation guide document still seem to be missing (from section Testing):
- Insert test_modeling_<model_type> into the list of tested modules in utils/run_tests.py.
- Append <model_type> to the list in check_adapters.py.
nice works! |
Followed the pattern of Bart to add adapters to T5. One change is that whereas Bart has separate classes for encoder and decoder, T5 does not. So I am using the
is_decoder
for changes between encoder and decoder classes, such as adding cross_attention adapters and adding invertible adapters.I'm working on some testing.