Back to Arena

Think-in-Memory

by Ant Group / Alibaba (Liu et al.)

System Card

OrganizationAnt Group / Alibaba (Liu et al.)
Released2023-11
Architectureepisodic-buffer / Post-thinking thought cache with LSH retrieval
DetailsAgent recalls relevant thoughts before responding, then post-thinks and writes the new thoughts back. Thoughts are organized via insert/forget/merge operations and retrieved with Locality-Sensitive Hashing, eliminating repeated biased reasoning.
Parameters—
Domainagent-memoryepisodic-session
Open SourceNo
PaperView Paper
lshpost-thinkingthoughtsevolved-memory

Capability Profile

Benchmark Scores

6 of 14 benchmarks
Long-Context Retrieval
0/5
RULER
no data
NIAH
no data
LooGLE
no data
LongBench
no data
∞Bench
no data
Multi-Turn Recall
2/2
LoCoMo
75.552p
MemoryBank
78.873p
Cross-Session Memory
1/1
LongMemEval
76.856p
Multi-Hop QA
2/3
BABILong
no data
MultiHop-RAG
70.439p
HotpotQA
70.145p
Agent Task Memory
1/1
AgentBench-Mem
7226p
Personalization
0/1
PerLTQA
no data
Factuality / Grounding
0/1
RAGAS
no data
Sources:Think-in-Memory paper (arXiv:2311.08719); evaluated on AgentBench Memory Track (Tsinghua KEG, 2308)Think-in-Memory paper (arXiv:2311.08719); evaluated on LoCoMo: Long-Term Conversational Memory Benchmark (Snap Research, 2402)Think-in-Memory paper (arXiv:2311.08719); evaluated on LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory (Salesforce AI Research, 2410)Think-in-Memory paper (arXiv:2311.08719); evaluated on HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering (Stanford / CMU, 1809)Think-in-Memory paper (arXiv:2311.08719); evaluated on MemoryBank: Enhancing LLMs with Long-Term Memory (Sun Yat-sen University, 2305)Think-in-Memory paper (arXiv:2311.08719); evaluated on MultiHop-RAG: Benchmarking Retrieval-Augmented Generation for Multi-Hop Queries (HKUST, 2401)