T2Mem first extracts lightweight episode-level metadata offline, capturing salient events and temporal associations as retrieval cues while keeping backlinks to raw interactions.
Building LLM memory at test time beats offline memory systems for long histories
The T2Mem framework interleaves retrieval and reasoning, progressively assembling query-specific memory from episode-level metadata.
Chinese Tech
Zijun Wang · Yifan Ji · Haidong Xin · Zhenghao Liu · Yukun Yan · Zulong Chen · +3 more
Northeastern University · Tsinghua University · Alibaba Group
Research Digest··2 min read
Wang et al.
Why this paper
From Alibaba Group and 2 others
In one line
Deferring memory construction to inference and building it progressively during reasoning improves long-context question answering with lower overhead.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§