Entity · paper

Copy Less, Ground More: Overcoming Repetitive Copying in Long-Context Reasoning via Evidence-Aware Reinforcement Learning

paperactiveprovisional

copy-less-ground-more-overcoming-repetitive-copying-in-long-context-reasoning-via-evidence-aware-reinforcement-learning-79f22dc3

·1 events·first seen 2d ago

Aliases: Copy Less, Ground More: Overcoming Repetitive Copying in Long-Context Reasoning via Evidence-Aware Reinforcement Learning

Co-occurring entities

GEAR

More like this (12)

Learning to Reason by Analogy via Retrieval-Augmented Reinforcement Fine-Tuning Language-Critique Imitation Learning from Suboptimal Demonstrations Difference-Aware Retrieval Policies for Imitation Learning Reinforcement Learning for Code Knowledgeless Language Models: Suppressing Parametric Recall for Evidence-Grounded Language Modeling RREDCoT: Segment-Level Reward Redistribution for Reasoning Models Bridging the Gap Between Latent and Explicit Reasoning with Looped Transformers Grounding LLM Reasoning under Incomplete Graph Evidence Long-context Reasoning Benchmarks Visually Grounded Self-Reflection for Vision-Language Models via Reinforcement Learning Invariant Learning Dynamics of Transformers in Inductive Reasoning Tasks Invariant Learning Dynamics of Transformers in Inductive Reasoning Tasks

Recent events (1)

6arXiv · cs.AI·2d ago·source ↗

GEAR: Reward shaping method reduces repetitive copying in long-context LLM reasoning

Researchers identify 'repetitive copying' as a pervasive failure mode in long-context LLMs, where models copy input text into reasoning traces instead of solving problems, worsening with context length. They propose GEAR (Grounding Evidence-Aware Reward), a reinforcement learning reward shaping method that adds a grounding reward for overlap with key evidence and a penalty for overlap with irrelevant distractor context. GEAR shows consistent improvements of up to +4.6 average points over standard accuracy-based RL across multiple model scales and benchmarks, with larger gains at longer contexts and reduced reasoning trace length. The work also introduces an automated pipeline for constructing evidence-annotated training data from arbitrary documents.

Long Context Evolution Evaluation and Benchmarking GEAR Copy Less, Ground More: Overcoming Repetitive Copying in Long-Context Reasoning via Evidence-Aware Reinforcement Learning +1 more