Mini-Sequence Transformers: Optimizing Intermediate Memory for Long Sequences Training
Cheng Luo, Jiawei Zhao, Zhuoming Chen, Beidi Chen, Anima Anandkumar
Abstract
We introduce Mini-Sequence Transformer (MsT), a simple and effective methodology for highly efficient and accurate LLM training with extremely long sequences. MsT partitions input sequences and iteratively processes mini-sequences to reduce intermediate memory usage. Integrated with activation recomputation, it enables significant memory savings in both forward and backward passes. In experiments with the Llama3-8B model, with MsT, we measure no degradation in throughput or convergence even with 12x longer sequences than standard implementations. MsT is fully general, implementation-agnostic, and requires minimal code changes to integrate with existing LLM training frameworks. Integrated with the huggingface library, MsT successfully extends the maximum context length of Qwen, Mistral, and Gemma-2 by 12-24x.
BibTeX
@inproceedings{
luo2024minisequence,
title={Mini-Sequence Transformers: Optimizing Intermediate Memory for Long Sequences Training},
author={Cheng Luo and Jiawei Zhao and Zhuoming Chen and Beidi Chen and Anima Anandkumar},
booktitle={The Thirty-eighth Annual Conference on Neural Information Processing Systems},
year={2024},
url={https://openreview.net/forum?id=2KuZHYykkq}
}