There’s a few *Trainer objects available from transformers, trl and setfit.
Other than the standard answer of “it depends on the task and which library you want to use”, what is the best practice or general guidelines when choosing which *Trainer object to use to train/tune our models?
Together with the *Trainer object, sometimes we see suggestions to use *TrainingArguments or the vanilla TrainingArguments.
For reference, we have:
- fine-tuning llama,
- https://github.com/philschmid/huggingface-llama-2-samples/blob/master/training/scripts/run_clm.py suggests vanilla
Trainer
- https://github.com/philschmid/huggingface-llama-2-samples/blob/master/training/scripts/run_clm.py suggests vanilla
- few-shot learning,
- SetFit: Efficient Few-Shot Learning Without Prompts suggests
from setfit import SetFitTrainer
- SetFit: Efficient Few-Shot Learning Without Prompts suggests
- machine translation
- reinforcement learning
- Supervised Fine-tuning Trainer suggests
from trl import SFTTrainer
- Supervised Fine-tuning Trainer suggests