Fine-tuning Quantized Neural Networks with Zeroth-order Optimization

• URL: http://arxiv.org/abs/2505.13430v1
• Date: Mon, 19 May 2025 17:55:15 GMT
• Title: Fine-tuning Quantized Neural Networks with Zeroth-order Optimization
• Authors: Sifeng Shang, Jiayi Zhou, Chenyu Lin, Minxian Li, Kaiyang Zhou,
• Abstract summary: Quantized Zeroth-order Optimization (QZO) is a novel approach that perturbs the continuous quantization scale for estimation and uses a directional derivative clipping method to stabilize training.<n>QZO can reduce the total memory cost by more than 18$times$ for 4-bit LLMs, and enables fine-tuning Llama-2-13B and Stable Diffusion 3.5 Large within a single 24GB GPU.
• Score: 18.645267970472936
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: As the size of large language models grows exponentially, GPU memory has become a bottleneck for adapting these models to downstream tasks. In this paper, we aim to push the limits of memory-efficient training by minimizing memory usage on model weights, gradients, and optimizer states, within a unified framework. Our idea is to eliminate both gradients and optimizer states using zeroth-order optimization, which approximates gradients by perturbing weights during forward passes to identify gradient directions. To minimize memory usage on weights, we employ model quantization, e.g., converting from bfloat16 to int4. However, directly applying zeroth-order optimization to quantized weights is infeasible due to the precision gap between discrete weights and continuous gradients, which would otherwise require de-quantization and re-quantization. To overcome this challenge, we propose Quantized Zeroth-order Optimization (QZO), a novel approach that perturbs the continuous quantization scale for gradient estimation and uses a directional derivative clipping method to stabilize training. QZO is orthogonal to both scalar-based and codebook-based post-training quantization methods. Compared to full-parameter fine-tuning in bfloat16, QZO can reduce the total memory cost by more than 18$\times$ for 4-bit LLMs, and enables fine-tuning Llama-2-13B and Stable Diffusion 3.5 Large within a single 24GB GPU.

Related papers

• Improving Block-Wise LLM Quantization by 4-bit Block-Wise Optimal Float (BOF4): Analysis and Variations [22.127873567034825]
  Large language models (LLMs) demand extensive memory capacity during both fine-tuning and inference.<n>Existing methods apply block-wise quantization techniques, such as NF4 and AF4, to the network weights.<n>We show that these quantization techniques incur suboptimal quantization errors.
  arXiv  Detail & Related papers  (2025-05-10T14:00:15Z)
• Pushing the Limits of Low-Bit Optimizers: A Focus on EMA Dynamics [64.62231094774211]
  Statefuls (e.g., Adam) maintain auxiliary information even 2x the model size in order to achieve optimal convergence.<n>SOLO enables Adam-styles to maintain quantized states with precision as low as 3 bits, or even 2 bits.<n>SOLO can thus be seamlessly applied to Adam-styles, leading to substantial memory savings with minimal accuracy loss.
  arXiv  Detail & Related papers  (2025-05-01T06:47:45Z)
• KurTail : Kurtosis-based LLM Quantization [51.24081396305435]
  KurTail is a new post-training quantization scheme that mitigates outliers in the activations of large language models.<n>It offers a 13.3% boost in MMLU accuracy and a 15.5% drop in Wiki perplexity compared to QuaRot.<n>It also outperforms SpinQuant with a 2.6% MMLU gain and reduces perplexity by 2.9%, all while reducing the training cost.
  arXiv  Detail & Related papers  (2025-03-03T12:43:06Z)
• QuZO: Quantized Zeroth-Order Fine-Tuning for Large Language Models [27.730213115659986]
  Language Models (LLMs) are often quantized to lower precision to reduce the memory cost and latency in inference.<n>Traditional fine-tuning methods require backpropagation, which are error-prone in the low-precision settings.<n>We propose the Quantized Zeroth-Order (QuZO) framework, specifically designed for fine-tuning LLMs through low-precision forward passes.
  arXiv  Detail & Related papers  (2025-02-17T22:20:31Z)
• SVDQuant: Absorbing Outliers by Low-Rank Components for 4-Bit Diffusion Models [58.5019443418822]
  Diffusion models can generate high-quality images, but as they scale, rising memory demands and higher latency pose deployment challenges.<n>We propose SVDQuant, a new 4-bit quantization paradigm to overcome this limitation.<n>We reduce the memory usage for the 12B FLUX.1 models by 3.5$times$, achieving 3.0$times$ speedup over the 4-bit weight-only quantization (W4A16) baseline.
  arXiv  Detail & Related papers  (2024-11-07T18:59:58Z)
• FlatQuant: Flatness Matters for LLM Quantization [58.28221892035609]
  We propose FlatQuant, a new post-training quantization approach that enhances the flatness of weights and activations.<n>Our approach identifies optimal affine transformations for each linear layer, calibrated in hours via a lightweight objective.<n>It achieves less than 1% accuracy drop for W4A4 quantization on the LLaMA-3-70B model, surpassing SpinQuant by 7.5%.
  arXiv  Detail & Related papers  (2024-10-12T08:10:28Z)
• Grass: Compute Efficient Low-Memory LLM Training with Structured Sparse Gradients [24.58231358634904]
  Large language model (LLM) training and finetuning are often bottlenecked by limited GPU memory. We propose Grass (GRAdient Stuctured Sparsification), a novel approach that leverages sparse projections to transform gradients into structured sparse updates.
  arXiv  Detail & Related papers  (2024-06-25T15:50:32Z)
• Winner-Take-All Column Row Sampling for Memory Efficient Adaptation of Language Model [89.8764435351222]
  We propose a new family of unbiased estimators called WTA-CRS, for matrix production with reduced variance. Our work provides both theoretical and experimental evidence that, in the context of tuning transformers, our proposed estimators exhibit lower variance compared to existing ones.
  arXiv  Detail & Related papers  (2023-05-24T15:52:08Z)
• LG-LSQ: Learned Gradient Linear Symmetric Quantization [3.6816597150770387]
  Deep neural networks with lower precision weights have advantages in terms of the cost of memory space and accelerator power. The main challenge associated with the quantization algorithm is maintaining accuracy at low bit-widths. We propose learned gradient linear symmetric quantization (LG-LSQ) as a method for quantizing weights and activation functions to low bit-widths.
  arXiv  Detail & Related papers  (2022-02-18T03:38:12Z)
• 8-bit Optimizers via Block-wise Quantization [57.25800395197516]
  Statefuls maintain statistics over time, e.g., the exponentially smoothed sum (SGD with momentum) or squared sum (Adam) of past values. This state can be used to accelerate optimization compared to plain gradient descent but uses memory that might otherwise be allocated to model parameters. In this paper, we develop first gradients that use 8-bit statistics while maintaining the performance levels of using 32-bit gradient states.
  arXiv  Detail & Related papers  (2021-10-06T15:43:20Z)

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.