Back to Arena

Recurrent Memory Transformer

by MIPT / DeepPavlov (Bulatov, Kuratov, Burtsev)

System Card

OrganizationMIPT / DeepPavlov (Bulatov, Kuratov, Burtsev)
Released2022-07
Architectureexternal-memory-network / Memory tokens passed between segments recurrently
DetailsAdds special memory tokens to each segment that pass information recurrently across segments of a long sequence, with no architectural changes beyond token-level memory slots.
Parameters—
Domainlong-context
Open SourceYes
PaperView Paper
CodeRepository
neurips-2022aaai-2024recurrentmemory-tokens1m-tokens

Capability Profile

Benchmark Scores

6 of 14 benchmarks
Long-Context Retrieval
5/5
RULER
79.594p
NIAH
75.962p
LooGLE
7977p
LongBench
603p
∞Bench
77.841p
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
80.390p
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:Recurrent Memory Transformer paper (arXiv:2207.06881); evaluated on BABILong: Testing the Limits of LLMs with Long-Context Reasoning-in-a-Haystack (AIRI, 2406)Recurrent Memory Transformer paper (arXiv:2207.06881); evaluated on InfiniteBench: Extending Long Context Evaluation Beyond 100K Tokens (Tsinghua / OpenBMB, 2402)Recurrent Memory Transformer paper (arXiv:2207.06881); evaluated on LongBench: A Bilingual, Multitask Benchmark for Long Context Understanding (Tsinghua KEG, 2308)Recurrent Memory Transformer paper (arXiv:2207.06881); evaluated on LooGLE: Can Long-Context Language Models Understand Long Contexts? (Peking University, 2311)Recurrent Memory Transformer paper (arXiv:2207.06881); evaluated on Needle in a Haystack (Greg Kamradt, 2024)Recurrent Memory Transformer paper (arXiv:2207.06881); evaluated on RULER: What's the Real Context Size of Your Long-Context Language Models (NVIDIA, 2404)