Multi-Turn Reasoning When Context Arrives in Pieces: Scalable Sharding and Memory-Augmented RL
For LLM practitioners, this work addresses the practical problem of conversational degradation with a scalable training method that improves incremental reasoning robustness.
LLM accuracy drops up to 65% in multi-turn conversations where context arrives piece by piece. A memory-augmented policy trained on sharded GSM8K significantly improves multi-turn accuracy and generalizes zero-shot to harder math and out-of-domain long-context QA, outperforming full-history baselines even when full history is available.
When a user reveals task-critical information across several conversation turns, LLM accuracy drops by up to 65% despite full context availability. We show that this Lost in Conversation degradation can be substantially mitigated by training models to maintain a compact rolling memory instead of attending to a growing history. To make such training scalable, we introduce a low-cost sharding pipeline that converts single-turn QA datasets into multi-turn fragmented-information episodes, eliminating the need for hours of manual annotation. Training only on sharded GSM8K, our memory-augmented policy significantly improves multi-turn accuracy and generalises zero-shot to harder math and out-of-domain long-context QA. Moreover, memory-trained models outperform full-history baselines even when given the full history at test time, suggesting that learning to compress induces more robust incremental reasoning than full-context exposure alone.