Research map / Architectures and efficiency / Pruning and efficient tuning
AGZO: Activation-Guided Zeroth-Order Optimization for LLM Fine-Tuning
Wei Lin, Yining Jiang, Qingyu Song, Qiao Xiang, Hong Xu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use AGZO activation-guided perturbations instead of isotropic MeZO | MeZO · CB (Qwen3-0.6B) | accuracy: 0.803 -> 0.892 | Improved |
| use AGZO instead of MeZO | MeZO · SST2 (Qwen3-0.6B) | accuracy: 0.858 -> 0.877 | Improved |
| use AGZO instead of MeZO | MeZO · BoolQ (Qwen3-0.6B) | accuracy: 0.730 -> 0.724 | Hurt |
| use AGZO instead of LOZO | LOZO · RTE (Qwen3-0.6B) | accuracy: 0.743 -> 0.772 | Improved |
| use AGZO instead of MeZO | MeZO · SST2 (Qwen3-4B) | accuracy: 0.875 -> 0.892 | Improved |
| use AGZO instead of MeZO | MeZO · BoolQ (Qwen3-4B) | accuracy: 0.823 -> 0.820 | No clear effect |
| use AGZO instead of MeZO | MeZO · MultiRC (Qwen3-4B) | accuracy: 0.850 -> 0.853 | No clear effect |
| use AGZO instead of MeZO | MeZO · Pangu-1B SST2 (GPU) | accuracy: 0.764 -> 0.778 | Improved |
| use AGZO instead of MeZO | MeZO · Pangu-1B CB (GPU) | accuracy: 0.732 -> 0.732 | No clear effect |
| use AGZO instead of MeZO | MeZO · Pangu-1B NPU average | accuracy: 0.703 -> 0.709 | Improved |
| use AGZO instead of MeZO | MeZO · Pangu-1B SST2 (NPU) | accuracy: 0.766 -> 0.765 | No clear effect |
| use AGZO instead of MeZO | MeZO · Pangu-1B CB (NPU) | accuracy: 0.732 -> 0.696 | Hurt |
Closest papers by meaning
- Enhancing Zeroth-order Fine-tuning for Language Models with Low-rank Structures (ICLR 2025 · similarity 0.90)
- Sparse MeZO: Less Parameters for Better Performance in Zeroth-Order LLM Fine-Tuning (NeurIPS 2025 · similarity 0.90)
- Trust the Direction, Search the Step: Zero-and-First-Order Methods for LLM Fine-Tuning (NeurIPS 2026 · similarity 0.89)
- Dominant-Layer ZO: A Single Layer Dominates Zeroth-Order Fine-Tuning of LLMs (NeurIPS 2026 · similarity 0.88)
- Revisiting Zeroth-Order Optimization for Memory-Efficient LLM Fine-Tuning: A Benchmark (ICML 2024 · similarity 0.88)
- Harmony in Divergence: Towards Fast, Accurate, and Memory-efficient Zeroth-order LLM Fine-tuning (NeurIPS 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
