Back to Arena

Selfmem

by Tsinghua / Microsoft (Cheng et al.)

System Card

OrganizationTsinghua / Microsoft (Cheng et al.)
Released2023-05
Architecturevector-rag / Iterative self-retrieval (generator feeds its own memory)
DetailsRetrieval-augmented generator iteratively produces outputs that a memory selector then chooses as retrieval targets for the next round, forming an unbounded self-memory pool rather than a fixed corpus.
Parameters—
Domainrag-retrieval
Open SourceYes
PaperView Paper
CodeRepository
neurips-2023self-retrievaltranslationsummarization

Capability Profile

Benchmark Scores

5 of 14 benchmarks
Long-Context Retrieval
2/5
RULER
68.424p
NIAH
no data
LooGLE
no data
LongBench
603p
∞Bench
no data
Multi-Turn Recall
0/2
LoCoMo
no data
MemoryBank
no data
Cross-Session Memory
0/1
LongMemEval
no data
Multi-Hop QA
2/3
BABILong
no data
MultiHop-RAG
6931p
HotpotQA
62.422p
Agent Task Memory
0/1
AgentBench-Mem
no data
Personalization
0/1
PerLTQA
no data
Factuality / Grounding
1/1
RAGAS
67.136p
Sources:Selfmem paper (arXiv:2305.02437); evaluated on HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering (Stanford / CMU, 1809)Selfmem paper (arXiv:2305.02437); evaluated on LongBench: A Bilingual, Multitask Benchmark for Long Context Understanding (Tsinghua KEG, 2308)Selfmem paper (arXiv:2305.02437); evaluated on MultiHop-RAG: Benchmarking Retrieval-Augmented Generation for Multi-Hop Queries (HKUST, 2401)Selfmem paper (arXiv:2305.02437); evaluated on RAGAS: Automated Evaluation of Retrieval-Augmented Generation (Exploding Gradients, 2309)Selfmem paper (arXiv:2305.02437); evaluated on RULER: What's the Real Context Size of Your Long-Context Language Models (NVIDIA, 2404)