AI Paper Atlas

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

ContextIF: Enhancing Instruction-Following through Context Reward

Yule Zhong, Jiacheng Yao, Guoxiu He

ICLR 2026 · Large language models › Instruction and fine-tuning · approach: Fine-tuning for reasoning data

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
add ContextIF generated contextLLaMA3-8B-Instruct · IFEvalIFEval avg: 77.11 -> 83.35Improved
add ContextIF generated contextLLaMA3-8B-Instruct · LiveBenchscore: 46.70 -> 59.90Improved
replace SPAR-8B with ContextIF-8BSPAR-8B · Multi-IF turn 3accuracy: 51.32 -> 53.51Improved
use randomly selected demonstrations as contextLLaMA3-8B-Instruct · IFEvalIFEval avg: 77.11 -> 76.97Hurt
generate context with GPT-4o instead of with RL-trained generatorContextIF-8B · IFEvalIFEval avg: 83.35 -> 82.77Hurt
fine-tune generator with SFT on 400 examples instead of RLLLaMA3-8B-Instruct LLM-context · IFEvalIFEval avg: 80.97 -> 80.34Hurt
add format reward (ablation)ContextIF-8B · IFEvalIFEval avg: 81.13 -> 83.35Improved
add summary reward (ablation)ContextIF-8B · IFEvalIFEval avg: 79.22 -> 83.35Improved
add demonstration question reward (ablation)ContextIF-8B · IFEvalIFEval avg: 81.15 -> 83.35Improved
add demonstration answer reward (ablation)ContextIF-8B · IFEvalIFEval avg: 78.50 -> 83.35Improved
swap judge LLaMA3-70B for Qwen-2.5-72B (ablation)ContextIF-8B · IFEvalIFEval avg: 83.35 -> 83.38No clear effect
add ContextIFLLaMA3-8B-Instruct · MMLUaccuracy: 68.40 -> 70.10Improved

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