Self-Alignment with Instruction Backtranslation
This addresses the challenge of efficiently aligning language models with human instructions for researchers and practitioners, though it is incremental as it builds on existing finetuning and data generation techniques.
The paper tackles the problem of building high-quality instruction-following language models by introducing instruction backtranslation, a scalable method that automatically labels human-written text with instructions through self-augmentation and self-curation. The result is a model that outperforms all other LLaMa-based models on the Alpaca leaderboard without using distillation data.
We present a scalable method to build a high quality instruction following language model by automatically labelling human-written text with corresponding instructions. Our approach, named instruction backtranslation, starts with a language model finetuned on a small amount of seed data, and a given web corpus. The seed model is used to construct training examples by generating instruction prompts for web documents (self-augmentation), and then selecting high quality examples from among these candidates (self-curation). This data is then used to finetune a stronger model. Finetuning LLaMa on two iterations of our approach yields a model that outperforms all other LLaMa-based models on the Alpaca leaderboard not relying on distillation data, demonstrating highly effective self-alignment.