# Addax: Utilizing Zeroth-Order Gradients to Improve Memory Efficiency and Performance of SGD for Fine-Tuning Language Models

arXiv: https://arxiv.org/abs/2410.06441 · HTML: https://mlsanity.com/paper/2410.06441/

## Key facts

- Cited by 11 papers in the map (data through 14 Oct 2026).
- Builds on 44 of its 70 references that are in the map.
- First posted on arXiv 9 Oct 2024.

## Builds on (most cited first, 30 of 44)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://mlsanity.com/paper/1810.04805/) (2018)
- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [PyTorch: An Imperative Style, High-Performance Deep Learning Library](https://mlsanity.com/paper/1912.01703/) (2019)
- [GPT-4 Technical Report](https://mlsanity.com/paper/2303.08774/) (2023)
- [LoRA: Low-Rank Adaptation of Large Language Models](https://mlsanity.com/paper/2106.09685/) (2021)
- [RoBERTa: A Robustly Optimized BERT Pretraining Approach](https://mlsanity.com/paper/1907.11692/) (2019)
- [Llama 2: Open Foundation and Fine-Tuned Chat Models](https://mlsanity.com/paper/2307.09288/) (2023)
- [SQuAD: 100,000+ Questions for Machine Comprehension of Text](https://mlsanity.com/paper/1606.05250/) (2016)
- [Prefix-Tuning: Optimizing Continuous Prompts for Generation](https://mlsanity.com/paper/2101.00190/) (2021)
- [The Power of Scale for Parameter-Efficient Prompt Tuning](https://mlsanity.com/paper/2104.08691/) (2021)
- [FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness](https://mlsanity.com/paper/2205.14135/) (2022)
- [OPT: Open Pre-trained Transformer Language Models](https://mlsanity.com/paper/2205.01068/) (2022)
- [QLoRA: Efficient Finetuning of Quantized LLMs](https://mlsanity.com/paper/2305.14314/) (2023)
- [A Broad-Coverage Challenge Corpus for Sentence Understanding through Inference](https://mlsanity.com/paper/1704.05426/) (2017)
- [ViLBERT: Pretraining Task-Agnostic Visiolinguistic Representations for Vision-and-Language Tasks](https://mlsanity.com/paper/1908.02265/) (2019)
- [BoolQ: Exploring the Surprising Difficulty of Natural Yes/No Questions](https://mlsanity.com/paper/1905.10044/) (2019)
- [A large annotated corpus for learning natural language inference](https://mlsanity.com/paper/1508.05326/) (2015)
- [SuperGLUE: A Stickier Benchmark for General-Purpose Language Understanding Systems](https://mlsanity.com/paper/1905.00537/) (2019)
- [Don't Stop Pretraining: Adapt Language Models to Domains and Tasks](https://mlsanity.com/paper/2004.10964/) (2020)
- [Making Pre-trained Language Models Better Few-shot Learners](https://mlsanity.com/paper/2012.15723/) (2020)
- [LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale](https://mlsanity.com/paper/2208.07339/) (2022)
- [Training Deep Nets with Sublinear Memory Cost](https://mlsanity.com/paper/1604.06174/) (2016)
- [Stochastic First- and Zeroth-order Methods for Nonconvex Stochastic Programming](https://mlsanity.com/paper/1309.5549/) (2013)
- [Intrinsic Dimensionality Explains the Effectiveness of Language Model Fine-Tuning](https://mlsanity.com/paper/2012.13255/) (2020)
- [GaLore: Memory-Efficient LLM Training by Gradient Low-Rank Projection](https://mlsanity.com/paper/2403.03507/) (2024)
- [WiC: the Word-in-Context Dataset for Evaluating Context-Sensitive Meaning Representations](https://mlsanity.com/paper/1808.09121/) (2018)
- [Measuring the Intrinsic Dimension of Objective Landscapes](https://mlsanity.com/paper/1804.08838/) (2018)
- [Fine-Tuning Language Models with Just Forward Passes](https://mlsanity.com/paper/2305.17333/) (2023)
- [PromptSource: An Integrated Development Environment and Repository for Natural Language Prompts](https://mlsanity.com/paper/2202.01279/) (2022)

## Built on by (most cited first, 11 of 11)

- [Zeroth-Order Optimization Finds Flat Minima](https://mlsanity.com/paper/2506.05454/) (2025)
- [MaZO: Masked Zeroth-Order Optimization for Multi-Task Fine-Tuning of Large Language Models](https://mlsanity.com/paper/2502.11513/) (2025)
- [SubZero: Composing Subject, Style, and Action via Zero-Shot Personalization](https://mlsanity.com/paper/2502.19673/) (2025)
- [Memory-Efficient Differentially Private Training with Gradient Random Projection](https://mlsanity.com/paper/2506.15588/) (2025)
- [Towards Fast LLM Fine-tuning through Zeroth-Order Optimization with Projected Gradient-Aligned Perturbations](https://mlsanity.com/paper/2510.18228/) (2025)
- [On the Optimal Construction of Unbiased Gradient Estimators for Zeroth-Order Optimization](https://mlsanity.com/paper/2510.19953/) (2025)
- [Private Zeroth-Order Optimization with Public Data](https://mlsanity.com/paper/2511.10859/) (2025)
- [VAMO: Efficient Zeroth-Order Variance Reduction for SGD with Faster Convergence](https://mlsanity.com/paper/2505.13954/) (2025)
- [Hi-ZFO: Hierarchical Zeroth- and First-Order LLM Fine-Tuning via Importance-Guided Tensor Selection](https://mlsanity.com/paper/2601.05501/) (2026)
- [ConMeZO: Adaptive Descent-Direction Sampling for Gradient-Free Finetuning of Large Language Models](https://mlsanity.com/paper/2511.02757/) (2025)
- [Unifying Data, Memory, and Compute Efficiency in LLM training: A Survey](https://mlsanity.com/paper/2606.10706/) (2026)

## Abstract (arXiv)

Fine-tuning language models (LMs) with the Adam optimizer often demands excessive memory, limiting accessibility. The "in-place" version of Stochastic Gradient Descent (IP-SGD) and Memory-Efficient Zeroth-order Optimizer (MeZO) have been proposed to address this. However, IP-SGD still requires substantial memory, and MeZO suffers from slow convergence and degraded final performance due to its zeroth-order nature. This paper introduces Addax, a novel method that improves both memory efficiency and performance of IP-SGD by integrating it with MeZO. Specifically, Addax computes zeroth- or first-order gradients of data points in the minibatch based on their memory consumption, combining these gradient estimates to update directions. By computing zeroth-order gradients for data points that require more memory and first-order gradients for others, Addax overcomes the slow convergence of MeZO and the excessive memory requirement of IP-SGD. Additionally, the zeroth-order gradient acts as a regularizer for the first-order gradient, further enhancing the model's final performance. Theoretically, we establish the convergence of Addax under mild assumptions, demonstrating faster convergence and less restrictive hyper-parameter choices than MeZO. Our experiments with diverse LMs and tasks show that Addax consistently outperforms MeZO regarding accuracy and convergence speed while having a comparable memory footprint. When fine-tuning OPT-13B with one A100 GPU, on average, Addax outperforms MeZO in accuracy/F1 score by 14% and runs 15x faster while using memory similar to MeZO. In our experiments on the larger OPT-30B model, on average, Addax outperforms MeZO in terms of accuracy/F1 score by >16 and runs 30x faster on a single H100 GPU. Moreover, Addax surpasses the performance of standard fine-tuning approaches, such as IP-SGD and Adam, in most tasks with significantly less memory requirement.