Better & Faster Large Language Models via Multi-token Prediction
Large language models, such as GPT and Llama, are trained with a next-token prediction loss.
In this work, they suggest that training language models to predict multiple future tokens at once, results in higher sample efficiency. More specifically, at each position in the training corpus, they ask the model to predict the following n tokens using n independent output heads, operating on top of a shared model trunk. Considering multi-token prediction as an auxiliary training task, they measure improved downstream capabilities, with no overhead in training time, for both code and natural language models.
This method is increasingly useful for larger model sizes and keeps its appeal when training for multiple epochs. Gains are especially pronounced on generative benchmarks like coding, where their models consistently outperform strong baselines by several percentage points.
The 13B parameter models solves 12 % more problems on HumanEval and 17 % more on MBPP than comparable next-token models. Experiments on small algorithmic tasks demonstrate that multi-token prediction is favorable for the development of induction heads and algorithmic reasoning capabilities.
As an additional benefit, models trained with 4-token prediction are up to 3× faster at inference, even with large batch sizes.
5
2 comments
Marcio Pacheco
7
Better & Faster Large Language Models via Multi-token Prediction
Data Alchemy
skool.com/data-alchemy
Your Community to Master the Fundamentals of Working with Data and AI — by Datalumina®
Leaderboard (30-day)
powered by