Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Gradient Transformer: Learning to Generate Updates for LLMs

About

Many organizations lack computational resources to fine-tune large language models (LLMs) on private (unshareable) data for better utility, while fine-tuning tiny language models (TinyLMs) alone performs poorly. To address this bottleneck, we propose a data-free knowledge distillation framework that generates LLM update vectors based on TinyLMs fine-tuned on private data. An update vector is a vector of parameter changes from an initial model to its fine-tuned version on a dataset, capturing the effect of cumulative gradient steps during fine-tuning. The key idea of our framework is a novel Gradient Transformer that transforms TinyLM's update vectors into LLM's update vectors. As derived from shadow datasets, Grad-Transformer captures the correlation between TinyLM and LLM update vectors, enabling third-party providers to generate LLM update vectors given the organization's TinyLM update vectors without accessing the organization's private data. The framework supports multi-organization collaboration to jointly update LLMs, improving performance and cost-efficiency. Extensive experiments across language modeling and reasoning tasks show that Grad-Transformer remarkably outperforms state-of-the-art knowledge distillation baselines, even under strict differential privacy protection.

Binh-Nguyen Nguyen, Khang Tran, NhatHai Phan, Issa Khalil• 2026

Related benchmarks

TaskDatasetResultRank
Commonsense ReasoningCommonsenseQA Multiple Client (test)
Client 1 Accuracy80.61
6
Mathematical ReasoningAQuA-RAT Single Client
Accuracy0.6102
6
Dialogue SummarizationSAMSum Single Client
ROUGE-150.52
6
Dialogue SummarizationDialogSum Single Client
ROUGE-148.37
6
Mathematical ReasoningGSM8K Single Client
Accuracy73.59
6
Mathematical ReasoningAQuA-RAT Multiple Client (test)
Client 1 Accuracy61.74
6
Reading ComprehensionDROP Single Client
Accuracy (DROP Single Client)58.26
6
Reading ComprehensionDROP Multiple Client (test)
Client 1 Accuracy60.72
6
Commonsense ReasoningCommonsenseQA Single Client
Accuracy83.21
6
Dialogue SummarizationSAMSum Multiple Client (test)
ROUGE-1 (Client 1)47.92
6
Showing 10 of 15 rows

Other info

Follow for update