Jesse Swanson
Jesse Swanson
@zphang @HaokunLiu Are T5 or GPT on your research roadmaps?
@HyeondeyKim Are you interested in integrating T5 with `jiant`? If so, I can sketch out how this would be done.
Only load shared encoder and do not load any of the heads. Also document different load modes.
By "more", what immediately comes to mind?
If `jiant` switches to completely use https://github.com/huggingface/nlp, `caching.py` will no longer be necessary.
**Comment by [sagnik](https://github.com/sagnik)** _Wednesday Sep 16, 2020 at 19:23 GMT_ ---- This is with `bert-base-uncased` if that's required.
**Comment by [sleepinyourhat](https://github.com/sleepinyourhat)** _Thursday Sep 17, 2020 at 16:59 GMT_ ---- @iftenney Is this familiar?
**Comment by [sleepinyourhat](https://github.com/sleepinyourhat)** _Thursday Sep 17, 2020 at 17:00 GMT_ ---- @iftenney (Alternately, we have rebuilt-from-scratch support for Edge Probing in v2 alpha, with a more active development/support team: https://github.com/jiant-dev/jiant...
**Comment by [sleepinyourhat](https://github.com/sleepinyourhat)** _Wednesday Sep 16, 2020 at 19:02 GMT_ ---- SQuAD: @zphang Do you there's a v1 branch with SQuAD support? Sadly, I fear that that reference may have...
**Comment by [sleepinyourhat](https://github.com/sleepinyourhat)** _Wednesday Jul 08, 2020 at 16:16 GMT_ ---- If you're looking at the raw data, I'm not sure we'll be able to help. You'd be better off...