DataDecide: How to Predict Best Pretraining Data with Small Experiments
This work addresses the high cost of pretraining large language models by enabling efficient dataset selection through small-scale experiments, which is an incremental improvement for researchers and practitioners in AI.
The paper tackles the problem of predicting the best pretraining datasets for large language models using small-scale experiments, finding that ranking models at a small size (e.g., 150M parameters) predicts larger-scale performance (1B) with about 80% accuracy and that continuous likelihood metrics in small experiments can predict benchmarks with over 80% accuracy using only 0.01% of the compute.
Because large language models are expensive to pretrain on different datasets, using smaller-scale experiments to decide on data is crucial for reducing costs. Which benchmarks and methods of making decisions from observed performance at small scale most accurately predict the datasets that yield the best large models? To empower open exploration of this question, we release models, data, and evaluations in DataDecide -- the most extensive open suite of models over differences in data and scale. We conduct controlled pretraining experiments across 25 corpora with differing sources, deduplication, and filtering up to 100B tokens, model sizes up to 1B parameters, and 3 random seeds. We find that the ranking of models at a single, small size (e.g., 150M parameters) is a strong baseline for predicting best models at our larger target scale (1B) (~80% of com parisons correct). No scaling law methods among 8 baselines exceed the compute-decision frontier of single-scale predictions, but DataDecide can measure improvement in future scaling laws. We also identify that using continuous likelihood metrics as proxies in small experiments makes benchmarks including MMLU, ARC, HellaSwag, MBPP, and HumanEval >80% predictable at the target 1B scale with just 0.01% of the compute.