一作ACL 2026Findings· Findings of ACL
Activation Reward Models for Few-Shot Model Alignment
Tianning Chai · Chancharik Mitra · Brandon Huang · Gautam Rajendrakumar Gare · Zhiqiu Lin · Assaf Arbelle · Leonid Karlinsky · Rogerio Feris · trevor darrell · Deva Ramanan · Roei Herzig
摘要
Aligning Large Language Models (LLMs) and Large Multimodal Models (LMMs) to human preferences is crucial for improving their real-world behavior. A common approach is to use reward models that enable reinforcement-learning post-training. However, traditional reward modeling requires finetuning on...