# Inference-Native Zeroth-Order Optimization for LLMs

arXiv: https://arxiv.org/abs/2605.28760 · HTML: https://mlsanity.com/paper/2605.28760/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 26 of its 36 references that are in the map.
- First posted on arXiv 27 May 2026.

## Builds on (most cited first, 26 of 26)

- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [Efficient Memory Management for Large Language Model Serving with PagedAttention](https://mlsanity.com/paper/2309.06180/) (2023)
- [GLUE: A Multi-Task Benchmark and Analysis Platform for Natural Language Understanding](https://mlsanity.com/paper/1804.07461/) (2018)
- [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)
- [HybridFlow: A Flexible and Efficient RLHF Framework](https://mlsanity.com/paper/2409.19256/) (2024)
- [GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers](https://mlsanity.com/paper/2210.17323/) (2022)
- [SuperGLUE: A Stickier Benchmark for General-Purpose Language Understanding Systems](https://mlsanity.com/paper/1905.00537/) (2019)
- [Fine-Tuning Language Models with Just Forward Passes](https://mlsanity.com/paper/2305.17333/) (2023)
- [FP8 Formats for Deep Learning](https://mlsanity.com/paper/2209.05433/) (2022)
- [Revisiting Zeroth-Order Optimization for Memory-Efficient LLM Fine-Tuning: A Benchmark](https://mlsanity.com/paper/2402.11592/) (2024)
- [Punica: Multi-Tenant LoRA Serving](https://mlsanity.com/paper/2310.18547/) (2023)
- [Sparse MeZO: Less Parameters for Better Performance in Zeroth-Order LLM Fine-Tuning](https://mlsanity.com/paper/2402.15751/) (2024)
- [Second-Order Fine-Tuning without Pain for LLMs:A Hessian Informed Zeroth-Order Optimizer](https://mlsanity.com/paper/2402.15173/) (2024)
- [Enhancing Zeroth-order Fine-tuning for Language Models with Low-rank Structures](https://mlsanity.com/paper/2410.07698/) (2024)
- [Evolution Strategies at Scale: LLM Fine-Tuning Beyond Reinforcement Learning](https://mlsanity.com/paper/2509.24372/) (2025)
- [DPZero: Private Fine-Tuning of Language Models without Backpropagation](https://mlsanity.com/paper/2310.09639/) (2023)
- [Evolution Strategies at the Hyperscale](https://mlsanity.com/paper/2511.16652/) (2025)
- [Powering Up Zeroth-Order Training via Subspace Gradient Orthogonalization](https://mlsanity.com/paper/2602.17155/) (2026)
- [HyGen: Efficient LLM Serving via Elastic Online-Offline Request Co-location](https://mlsanity.com/paper/2501.14808/) (2025)
- [FlexLLM: Token-Level Co-Serving of LLM Inference and Finetuning with SLO Guarantees](https://mlsanity.com/paper/2402.18789/) (2024)
- [Zeroth-Order Optimization is Secretly Single-Step Policy Optimization](https://mlsanity.com/paper/2506.14460/) (2025)
- [ZO2: Scalable Zeroth-Order Fine-Tuning for Extremely Large Language Models with Limited GPU Memory](https://mlsanity.com/paper/2503.12668/) (2025)
- [MobiZO: Enabling Efficient LLM Fine-Tuning at the Edge via Inference Engines](https://mlsanity.com/paper/2409.15520/) (2024)
- [AdaMeZO: Adam-style Zeroth-Order Optimizer for LLM Fine-tuning Without Maintaining the Moments](https://mlsanity.com/paper/2605.00650/) (2026)

## Abstract (arXiv)

Zeroth-order (ZO) methods train large language models using only forward passes, yet common ZO execution paths perform substantial work beyond what the algorithm itself requires. To remove this extra execution overhead, we present Infer-ZO, which separates the evaluations required by the algorithm from how they are executed, allowing them to reuse existing inference-engine optimizations. After removing inherited execution work, a complete Infer-ZO step on Qwen3-14B adds only 1.16% wall-clock time over matched inference, bringing ZO execution close to its fundamental inference workload. With Infer-ZO, ZO evaluations run at near-inference cost with frozen base weights and can share GPU batches with ordinary inference requests. In co-serving experiments, foreground and background throughput remain within 1-3% of the corresponding background-request baseline. Across 15 Qwen3, Llama, and OPT models, Infer-ZO achieves 2.10x-9.94x end-to-end step speedups over released LoZO. Code is available at https://github.com/playeriv65/zo-vllm.