SUMONE summarizer compresses agent histories by 78% on DeepSearchQA
Original title: SUMONE: A Simple and Efficient Summarizer for Plug-and-Play Agent Memory
SUMONE summarizes trajectories via masked reconstruction and GRPO reward for task-agnostic plug-and-play compression.
Long-horizon language-model agents accumulate large interaction histories. SUMONE is a standalone summarizer trained to preserve information through masked reconstruction. It optimizes the summarizer using GRPO on raw trajectories and reduces estimated inference cost by approximately 78% on DeepSearchQA and 75% on SWE-bench Verified relative to full context.
- Reduces estimated inference cost by 78% on DeepSearchQA
- Reduces 75% on SWE-bench Verified
- Transfers across downstream models and agent harnesses without fine-tuning
- Trained with self-supervision from raw trajectories using GRPO
From the original article
1 min read at arxiv.orgLong-horizon language-model agents accumulate increasingly large interaction histories, making context compaction essential for efficient execution. Existing learned compaction methods often rely on downstream task outcomes or solver behavior, coupling training to specific tasks or agents. We introduce SumOne, a standalone summarizer trained to preserve information in agent histories through masked reconstruction.
By Jizhe Zhang · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.