Instruction Backtranslation Model
This model implements the backward model (Myx) from the "Self-Alignment with Instruction Backtranslation" paper.
It was trained on the OpenAssistant-Guanaco dataset to predict instructions from outputs.
Details
- Model trained on output-instruction pairs
- Implements part 1 of the paper: creating a model that predicts instructions from outputs (Myx)