Update modeling to keep up with phi-2
#11
by
ma2za
- opened
The microsoft people are doing a lot of changes to the modeling and config. Do you plan to keep working on this model? if so i can give you a hand given that i am already doing some changes to support beam search.
I don't have compute to retrain the model
I think you can merge the adapter model and use the script(https://github.com/huggingface/transformers/blob/main/src/transformers/models/phi/convert_phi_weights_to_hf.py) to convert it into the current modeling and config.
I did the conversion for mine: https://huggingface.co/vince62s/phi-2-psy
I'll try to do it next week for the two source model.