LLM Pretraining with Continuous Concepts

Jihoon Tack, Jack Lanchantin, Jane Yu, Andrew Cohen, Ilia Kulikov, Janice Lan, Shibo Hao, Yuandong Tian, Jason E Weston, Xian Li

Open Source Links: https://github.com/facebookresearch/RAM/tree/main/projects/cocomix

Keywords: Sparse Autoencoders, Other

Other Keywords: Large Language Models, Pretraining

TL;DR: We propose Continuous Concept Mixing, a pretraining framework that improves sample efficiency and interpretability by jointly predicting the next token and concepts from a sparse autoencoder, then mixing these concepts into the model's hidden states

Abstract: Next token prediction has been the standard training objective used in large language model pretraining. Representations are learned as a result of optimizing for token-level perplexity. We propose Continuous Concept Mixing (CoCoMix), a novel pretraining framework that combines discrete next token prediction with continuous concepts. Specifically, CoCoMix predicts ``continuous concepts'' learned from a pretrained sparse autoencoder and mixes them into the model's hidden state by interleaving with token hidden representations. Through experiments on multiple benchmarks, including language modeling and downstream reasoning tasks, we show that CoCoMix is more sample efficient and consistently outperforms standard next token prediction and knowledge distillation. We find that combining both concept learning and interleaving in an end-to-end framework is critical to performance gains. Furthermore, CoCoMix enhances interpretability and steerability by allowing direct inspection and modification of the predicted concept, offering a transparent way to guide the model’s internal reasoning process.

Submission Number: 9