Hi-ZFO: Hierarchical Zeroth- and First-Order LLM Fine-Tuning via Importance-Guided Tensor Selection
This work addresses the challenge of efficient and effective fine-tuning for LLMs, which is crucial for improving model generalization in various applications, though it appears incremental as it builds on existing optimization methods.
The paper tackled the problem of fine-tuning large language models (LLMs) by addressing the trade-offs between first-order (FO) optimization, which leads to sharp minima, and zeroth-order (ZO) methods, which are slow and noisy. It proposed Hi-ZFO, a hybrid framework that adaptively applies FO and ZO updates based on layer importance, achieving superior performance and reduced training time across generative, mathematical, and code reasoning tasks.
Fine-tuning large language models (LLMs) using standard first-order (FO) optimization often drives training toward sharp, poorly generalizing minima. Conversely, zeroth-order (ZO) methods offer stronger exploratory behavior without relying on explicit gradients, yet suffer from slow convergence. More critically, our analysis reveals that in generative tasks, the vast output and search space significantly amplify estimation variance, rendering ZO methods both noisy and inefficient. To address these challenges, we propose \textbf{Hi-ZFO} (\textbf{Hi}erarchical \textbf{Z}eroth- and \textbf{F}irst-\textbf{O}rder optimization), a hybrid framework designed to synergize the precision of FO gradients with the exploratory capability of ZO estimation. Hi-ZFO adaptively partitions the model through layer-wise importance profiling, applying precise FO updates to critical layers while leveraging ZO optimization for less sensitive ones. Notably, ZO in Hi-ZFO is not merely a memory-saving surrogate; it is intentionally introduced as a source of "beneficial stochasticity" to help the model escape the local minima where pure FO optimization tends to stagnate. Validated across diverse generative, mathematical, and code reasoning tasks, Hi-ZFO consistently achieves superior performance while significantly reducing the training time. These results demonstrate the effectiveness of hierarchical hybrid optimization for LLM fine-tuning.