Sample-Adaptive Zeroth-Order Post-Training
To address this limitation, we propose Sample-Adaptive Zeroth-Order Optimization (SAZO), which learns sample importance online and adaptively allocates queries to accelerate convergence.
ProofPaper ↗
Key points
- Zeroth-order (ZO) optimization offers a memory-efficient alternative to backpropagation for fine-tuning large language models.
- SAZO extracts gradient-norm information from existing directional differences to estimate sample importance without access to true gradients.
- We prove that SAZO asymptotically matches the leading convergence term of the minimum-variance sampling oracle, and achieves a strictly faster convergence rate than standard methods.
- Extensive experiments across four model families demonstrate improved average downstream performance and up to $6\times$ faster convergence to the same training loss in terms of query count.
Sources (1)
- [1]Sample-Adaptive Zeroth-Order Post-TrainingarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 8, 06:00 PM
To address this limitation, we propose Sample-Adaptive Zeroth-Order Optimization (SAZO), which learns sample importance online and adaptively allocates queries to accelerate convergence.
Zeroth-order (ZO) optimization offers a memory-efficient alternative to backpropagation for fine-tuning large language models.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 8, 2026Understanding Jev, the new model everyone is talking about
- Oct 8, 2026Beyond Imitation: A Framework and Benchmark for LLM-Assisted Peer Review
- Oct 8, 2026SuperNav: An Agentic Navigation System for Any Task in Any Scene
- Oct 8, 2026VibeEdit: Image Editing with Canvas Instructions
- Oct 8, 2026SpaceCast-Bench: Evaluating Predictive Spatial Reasoning in Vision-Language Models
- Oct 7, 2026Iris-3B: Going Beyond the Latent with Pixel-Space Diffusion Training, Conversion and Fine-Tuning
