The Pythia-160M model is designed for research on language model behavior and interpretability, trained on the Pile dataset. Here we've evaluated it on HELLASWAG and can be fine-tuned for further experimentation.

Evaluated on the Eleuther evaluation harness, revision 100,000 steps

Done just an exercise - not intended for deployment or human-facing interactions.