AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

Prompt and Parameter Co-Optimization for Large Language Models

Xiaohe Bo, Rui Li, Zexu Sun, Quanyu Dai, Zeyu Zhang, Zihang Tian, Xu Chen, Zhenhua Dong

ICLR 2026 · Large language models › Instruction and fine-tuning · approach: Fine-tuning generalization and delta compression

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 changedWhereMeasuredEffect
jointly generate prompts and LoRA parameters instead of BetterTogetherMetaTuner-J (Qwen2.5-7B) · MATHEM accuracy: 41.56 -> 48.67Improved
jointly generate prompts and LoRA parameters instead of BetterTogetherMetaTuner-J (Qwen2.5-3B) · HotpotQAF1: 41.10 -> 58.85Improved
hybrid optimization instead of fine-tuning onlyMetaTuner-J (Qwen2.5-7B) · GSM8KEM accuracy: 63.68 -> 78.92Improved
add fine-tuning term to MetaTuner (ablation)MetaTuner (Qwen2.5-7B) · MATHEM accuracy: 48.00 -> 48.67Improved
add prompt-optimization term to MetaTuner (ablation)MetaTuner (Qwen2.5-7B) · MATHEM accuracy: 46.22 -> 48.67Improved
share meta encoder between prompt and parameter decoders (ablation)MetaTuner (Qwen2.5-7B) · MATHEM accuracy: 46.67 -> 48.67Improved
share meta encoder (ablation)MetaTuner (Qwen2.5-7B) · CosmosQAEM accuracy: 91.50 -> 92.25Mixed
use sequential (-I) instead of joint (-J) optimizationMetaTuner · HotpotQAF1: 54.56 -> 55.75Mixed
replace supervised regularizer with Gumbel-Softmax (ablation)MetaTuner · benchmarksaccuracyHurt
generate D2 from the model's own rollouts instead of an expert LLM (ablation)MetaTuner · benchmarksaccuracyImproved
increase number of rollout samples (ablation)MetaTuner · benchmarksaccuracyHurt
update prompt decoder used for rollouts less frequently (ablation)MetaTuner · benchmarksaccuracyHurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy