Add support for Pegasus
Environment info
-
adapter-transformersversion: Latest of (Jan 2022) - Platform: Google colab
- Python version: Latest of (Jan 2022)
- PyTorch version (GPU?): -
Details
Hey there, I'm working with the PEGASUS model and want to add an adapter to it. I tried with EncoderDecoderModel by adding config of PegasusConfig but didn't work as there is no module of Adapter in PegasusConfig as we have in other models like BartConfig, BertConfig, T5Config, and all. As per my understanding, I should modify the configuration of PEGASUS so that it can adapt the Adapter. Now the question is how?
Hoping for an effective solution.
Thanks :)
Hey @Darshan2104,
unfortunately, we currently don't support adapters for Pegasus models yet. Adapter support would have to be added to the modeling class first.
Hello@calpt Firstly Thanks for response! Is it possible that I can work on adding that Adapter support for Pegasus? If yes, can you just elaborate the flow on how can I do that? Thanks:)
Leaving this open as feature request for Pegasus support.
@Darshan2104 I'll ping you as soon as we're ready with the current changes if you're still interested in contributing to this :)
sure sure. I'm eager to do a contribution!!! @calpt It would be my honor to work with you!
Hey @Darshan2104, just wanted to let you know that v3 is now released and we're open for new model contributions. To get started, please have a look at our model contributing guide. Let us know if there's anything we can help with!
Hello @calpt, I'll start working on it within a week. I'll definitely text you for help whenever I get stuck somewhere. I'll also give you weekly or every 10day working progress. Thanks :)
Hello @calpt , I've started working on it, but I have a bit of confusion. How do I check whether whatever new file(i.e mixins) I'm creating is working well or not, is there any way of testing individual files?!?!
Hey @Darshan2104, unfortunately, there's no option to test invidividual changes directly while implementing them. However, we have an extensive suite of unit tests that check the full implementation. You can run these unit tests for the new model after following the implementation steps in the testing section of the model contribution guide.