When Less is More: 8-bit Quantization Improves Continual Learning in Large Language Models

• URL: http://arxiv.org/abs/2512.18934v1
• Date: Mon, 22 Dec 2025 00:51:39 GMT
• Title: When Less is More: 8-bit Quantization Improves Continual Learning in Large Language Models
• Authors: Michael S. Zhang, Rishi A. Ruia, Arnav Kewalram, Saathvik Dharmapuram, Utkarsh Sharma, Kevin Zhu,
• Abstract summary: We investigate interplay between quantization precision (FP16, INT8, INT4) and replay buffer strategies in large language models.<n>We observe a striking inversion on subsequent tasks: quantized models outperform FP16 by 8-15% on final task forward accuracy.<n>We hypothesize that quantization-induced noise acts as implicit regularization, preventing the overfitting to new task gradients that plagues high-precision models.
• Score: 3.395259076232696
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Catastrophic forgetting poses a fundamental challenge in continual learning, particularly when models are quantized for deployment efficiency. We systematically investigate the interplay between quantization precision (FP16, INT8, INT4) and replay buffer strategies in large language models, revealing unexpected dynamics. While FP16 achieves superior initial task performance (74.44% on NLU), we observe a striking inversion on subsequent tasks: quantized models outperform FP16 by 8-15% on final task forward accuracy, with INT4 achieving nearly double FP16's performance on Code generation (40% vs 20%). Critically, even minimal replay buffers (0.1%) dramatically improve retention - increasing NLU retention after Math training from 45% to 65% across all precision levels - with INT8 consistently achieving the optimal balance between learning plasticity and knowledge retention. We hypothesize that quantization-induced noise acts as implicit regularization, preventing the overfitting to new task gradients that plagues high-precision models. These findings challenge the conventional wisdom that higher precision is always preferable, suggesting instead that INT8 quantization offers both computational efficiency and superior continual learning dynamics. Our results provide practical guidelines for deploying compressed models in continual learning scenarios: small replay buffers (1-2%) suffice for NLU tasks, while Math and Code benefit from moderate buffers (5-10%), with quantized models requiring less replay than FP16 to achieve comparable retention. Code is available at https://github.com/Festyve/LessIsMore.

Related papers

• Pretraining Large Language Models with NVFP4 [53.235038214986865]
  We introduce a novel approach for stable and accurate training of large language models (LLMs) using the NVFP4 format.<n>Our method integrates two-dimensional quantization scheme for consistent representations across both the forward and backward passes.<n>Our results show that the model trained with our NVFP4-based pretraining technique achieves training loss and downstream task accuracies comparable to an FP8 baseline.
  arXiv  Detail & Related papers  (2025-09-29T17:53:17Z)
• Dual Precision Quantization for Efficient and Accurate Deep Neural Networks Inference [3.7687375904925484]
  We propose a novel hardware-efficient quantization and inference scheme that exploits hardware advantages with minimal accuracy degradation.<n>We develop a novel quantization algorithm, dubbed Dual Precision Quantization (DPQ), that leverages the unique structure of our scheme without introducing additional inference overhead.
  arXiv  Detail & Related papers  (2025-05-20T17:26:12Z)
• Optimizing Large Language Model Training Using FP4 Quantization [73.55459961002371]
  Quantized training presents a promising solution by enabling low-bit arithmetic operations to reduce costs.<n>This work introduces the first FP4 training framework for large language models (LLMs)
  arXiv  Detail & Related papers  (2025-01-28T18:04:50Z)
• "Give Me BF16 or Give Me Death"? Accuracy-Performance Trade-Offs in LLM Quantization [67.3213104337679]
  Quantization is a powerful tool for accelerating large language model (LLM) inference, but the accuracy-performance trade-offs across different formats remain unclear.<n>We conduct the most comprehensive empirical study to date, evaluating FP8, INT8, and INT4 quantization across academic benchmarks and real-world tasks.
  arXiv  Detail & Related papers  (2024-11-04T18:21:59Z)
• ZeroQuant-FP: A Leap Forward in LLMs Post-Training W4A8 Quantization Using Floating-Point Formats [25.543571445739936]
  This study explores the viability of floating-point (FP) quantization for large language models (LLMs) For LLMs, FP8 activation consistently outshines its integer (INT8) equivalent, with the performance edge becoming more noticeable in models possessing parameters beyond one billion. For weight quantization, our findings indicate that FP4 exhibits comparable, if not superior, performance to INT4, simplifying deployment on FP-supported hardware like H100.
  arXiv  Detail & Related papers  (2023-07-19T06:58:03Z)
• Quantized Neural Networks for Low-Precision Accumulation with Guaranteed Overflow Avoidance [68.8204255655161]
  We introduce a quantization-aware training algorithm that guarantees avoiding numerical overflow when reducing the precision of accumulators during inference. We evaluate our algorithm across multiple quantized models that we train for different tasks, showing that our approach can reduce the precision of accumulators while maintaining model accuracy with respect to a floating-point baseline.
  arXiv  Detail & Related papers  (2023-01-31T02:46:57Z)
• Make RepVGG Greater Again: A Quantization-aware Approach [22.36179771869403]
  We propose a simple, robust, and effective remedy to have a quantization-friendly structure. Without bells and whistles, the top-1 accuracy drop on ImageNet is reduced within 2% by standard post-training quantization. Our method also achieves similar FP32 performance as RepVGG.
  arXiv  Detail & Related papers  (2022-12-03T11:14:10Z)
• Q-ASR: Integer-only Zero-shot Quantization for Efficient Speech Recognition [65.7040645560855]
  We propose Q-ASR, an integer-only, zero-shot quantization scheme for ASR models. We show negligible WER change as compared to the full-precision baseline models. Q-ASR exhibits a large compression rate of more than 4x with small WER degradation.
  arXiv  Detail & Related papers  (2021-03-31T06:05:40Z)
• Towards Unified INT8 Training for Convolutional Neural Network [83.15673050981624]
  We build a unified 8-bit (INT8) training framework for common convolutional neural networks. First, we empirically find the four distinctive characteristics of gradients, which provide us insightful clues for gradient quantization. We propose two universal techniques, including Direction Sensitive Gradient Clipping that reduces the direction deviation of gradients.
  arXiv  Detail & Related papers  (2019-12-29T08:37:53Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.