Reversing Large Language Models for Efficient Training and Fine-Tuning

2025-11-27

AI Summary

Fine-tuning large language models (LLMs) incurs prohibitive memory overhead due to storing intermediate activations during forward propagation. Method: This paper proposes a reversible LLM architecture grounded in symmetric and symplectic differential equations, leveraging time-reversible dynamics to exactly reconstruct hidden states during backpropagation—eliminating the need to cache activations. Contribution/Results: It introduces the first reversible computing framework tailored for LLMs, featuring a plug-and-play reversible fine-tuning method that efficiently converts pretrained models without architectural modification. Evaluated across multiple mainstream LLMs and benchmark tasks, the approach achieves comparable or superior performance while reducing peak memory complexity from O(L) to O(1) (L = number of layers), enabling larger batch sizes, lower GPU memory consumption, and improved training throughput.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Computer Vision: Large Vision Models

Application Category

• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Large Language Models (LLMs) are known for their expensive and time-consuming training. Thus, oftentimes, LLMs are fine-tuned to address a specific task, given the pretrained weights of a pre-trained LLM considered a foundation model. In this work, we introduce memory-efficient, reversible architectures for LLMs, inspired by symmetric and symplectic differential equations, and investigate their theoretical properties. Different from standard, baseline architectures that store all intermediate activations, the proposed models use time-reversible dynamics to retrieve hidden states during backpropagation, relieving the need to store activations. This property allows for a drastic reduction in memory consumption, allowing for the processing of larger batch sizes for the same available memory, thereby offering improved throughput. In addition, we propose an efficient method for converting existing, non-reversible LLMs into reversible architectures through fine-tuning, rendering our approach practical for exploiting existing pre-trained models. Our results show comparable or improved performance on several datasets and benchmarks, on several LLMs, building a scalable and efficient path towards reducing the memory and computational costs associated with both training from scratch and fine-tuning of LLMs.

Problem

• Reducing memory consumption in LLM training
• Enabling larger batch sizes for improved throughput
• Converting non-reversible LLMs into reversible architectures

Innovation

• Reversible architectures reduce memory consumption
• Time-reversible dynamics retrieve hidden states
• Fine-tuning converts non-reversible models efficiently