Add modifications for the supervised fine-tuning aka warm startup for the scheme generator language model.
The code is working for T5 (enc-dec architecture). Considerations to make in the future:
If using decoder only LM, make use of causal LMs.
ensure that the complete target is taken into consideration while decoding.
Modifications
[x] Added DataPool
[x] Added script (currently uses 4 metrics for evaluation)
Description
Add modifications for the supervised fine-tuning aka warm startup for the scheme generator language model. The code is working for T5 (enc-dec architecture). Considerations to make in the future:
Modifications