Smaller Language Models are capable of selecting Instruction-Tuning Training Data for Larger Language Models
This offers a more efficient alternative for aligning language models, reducing training costs for practitioners, though it is incremental in data selection methods.
The paper tackles the high cost of instruction-tuning large language models by proposing that smaller models can autonomously select high-quality training data, resulting in comparable or improved performance; experiments show a 350M model curating data for a 13B model yields equal or superior results to using the full dataset.
Instruction-tuning language models has become a crucial step in aligning them for general use. Typically, this process involves extensive training on large datasets, incurring high training costs. In this paper, we introduce a novel training data selection based on the learning percentage of the samples. We assert that current language models possess the capability to autonomously select high-quality training data, leading to comparable or improved performance compared to training on the entire dataset. Our experiments span different-sized models, revealing that this characteristic holds for models ranging from 1B (small) to 13B (large) in size. Moreover, we demonstrate an interesting finding that the data hardness transfers across model sizes, and a smaller 350M model can effectively curate high-quality training data with hard samples for a larger 13B model, resulting in an equally or superior instruction-tuned model compared to training on the complete dataset. Utilizing open-sourced OPT and Llama-2 models up to 13B in size, two publicly available instruction-tuning training datasets and evaluated by both automatic metrics & humans, our paper introduces a novel approach to training data selection, showcasing a more efficient alternative.