Forked from https://github.com/LAION-AI/Open-Assistant.

Run example:

python \
   training/trainer.py \
   training/configs/multi/deberta-small-newsroom-inf.yml

Configs for multi training

Training configs allow to run multiple fine-tunings of the same base model on different splits. This is to allow training on 0%, 10%, ..., 100% of the training data, if the dataset is prepared to have training set split into subsets.

Eg. training/configs/multi/deberta-small-newsroom-inf.yml utilises the fact that for Newsroom we have training set split into train_0, train_1, ..., train_9, each containing ~10% of the training data. Each run is given a list of splits which are together used for training.