search

webis-de / set-encoder

13 stars 0 forks source link

Fully-trained models? #3

Closed sergeyf closed 5 months ago

sergeyf commented 5 months ago

Hello,

Where can the fully-trained 330M model be downloaded?

Also, if I wanted to train a larger model using your pipeline - is that currently supported in the code?

fschlatt commented 5 months ago

Hey!

I haven't gotten around to uploading the already fine-tuned models. I promise to do so by the end of the week ;) A fine-tuned monoELECTRA model is already uploaded, but unfortunately, how to use it is also not well documented. I will try to document everything ASAP.

Fine-tuning larger models should be doable, but currently only encoder-only bert-based backbones (bert, electra, and roberta) are supported. I'm not aware of any larger models bert-based encoder models larger than 330M parameters, but haven't looked either

sergeyf commented 5 months ago

Thanks!

fschlatt commented 5 months ago

I've uploaded the models and added a bit of info in the README. More detailed documentation will follow in the lightning-ir repository within the next few days

sergeyf commented 5 months ago

Wonderful, thanks!