Allows generation parameters to be set in a hydra config, and consequently be passed to an instance of GenerationConfig and further into the model.generate call
Implements early stopping of generation when certain token sequences are generated. This helps prevent base models from generating new Q-A pairs for qa datasets.
This PR does the following: