Dominant-Layer ZO: A Single Layer Dominates Zeroth-Order Fine-Tuning of LLMs
For practitioners of memory-efficient LLM fine-tuning, this reveals that ZO optimization can be drastically accelerated by focusing on a single layer, with no loss in performance.
Zeroth-order (ZO) fine-tuning of LLMs is dominated by a single decoding layer; fine-tuning only that layer matches or exceeds full-model ZO fine-tuning, achieving up to 4.52× training speedup on LLaMA2-7B and Qwen3-8B across nine benchmarks.
Zeroth-order (ZO) optimization enables memory-efficient fine-tuning of large language models (LLMs) using only forward passes, but it remains unclear how useful adaptation is distributed across layers. In this work, we reveal a surprising phenomenon: ZO fine-tuning is sharply dominated by a single decoding layer. Across multiple LLM families and downstream tasks, fine-tuning this dominant layer alone consistently matches or even exceeds full-model ZO fine-tuning. We further show that the dominant layer is task-agnostic but model-specific, and can be identified before training through a simple inference-only analysis of activation outliers. Specifically, the dominant layer consistently aligns with the first activation-outlier layer in the pre-trained model. To explain this phenomenon, we analyze how perturbation effects propagate under ZO optimization. We find that the dominant layer combines two key properties: high perturbation sensitivity and early placement in the residual stream, allowing perturbation-induced effects to propagate and accumulate through remaining subsequent decoding layers. As a result, this layer produces disproportionately strong and stable optimization signals under forward-only updates. Extensive experiments on LLaMA2-7B and Qwen3-8B across nine benchmarks show that dominant-layer ZO fine-tuning improves average performance over full-model MeZO and LoRA-based ZO fine-tuning while achieving up to 4.52$\times$ training speedup.