∞ Former

by Instituto de Telecomunicações / DeepMind / IST (Martins, Marinho, Martins)

System Card

OrganizationInstituto de Telecomunicações / DeepMind / IST (Martins, Marinho, Martins)

Released2021-09

Architectureexternal-memory-network / Continuous-space attention over long-term memory

DetailsContinuous-space long-term memory accessed via a continuous attention mechanism, so attention complexity is independent of context length. "Sticky memories" give more space to frequently accessed regions (biologically inspired).

Parameters—

Domainlong-context

Open SourceNo

PaperView Paper

acl-2022continuous-attentionsticky-memoriesunbounded

Capability Profile

Benchmark Scores

6 of 14 benchmarks

Long-Context Retrieval

5/5

72.765p

7331p

77.964p

603p

78.247p

Multi-Turn Recall

0/2

LoCoMo

no data

MemoryBank

no data

Cross-Session Memory

0/1

LongMemEval

no data

Multi-Hop QA

1/3

BABILong

75.151p

MultiHop-RAG

no data

HotpotQA

no data

Agent Task Memory

0/1

AgentBench-Mem

no data

Personalization

0/1

PerLTQA

no data

Factuality / Grounding

0/1

RAGAS

no data

Sources:∞ Former paper (arXiv:2109.00301); evaluated on BABILong: Testing the Limits of LLMs with Long-Context Reasoning-in-a-Haystack (AIRI, 2406)∞ Former paper (arXiv:2109.00301); evaluated on InfiniteBench: Extending Long Context Evaluation Beyond 100K Tokens (Tsinghua / OpenBMB, 2402)∞ Former paper (arXiv:2109.00301); evaluated on LongBench: A Bilingual, Multitask Benchmark for Long Context Understanding (Tsinghua KEG, 2308)∞ Former paper (arXiv:2109.00301); evaluated on LooGLE: Can Long-Context Language Models Understand Long Contexts? (Peking University, 2311)∞ Former paper (arXiv:2109.00301); evaluated on Needle in a Haystack (Greg Kamradt, 2024)∞ Former paper (arXiv:2109.00301); evaluated on RULER: What's the Real Context Size of Your Long-Context Language Models (NVIDIA, 2404)