Back to Arena

kNN-LM

by Stanford / Facebook AI Research (Khandelwal et al.)

System Card

OrganizationStanford / Facebook AI Research (Khandelwal et al.)
Released2019-11
Architectureexternal-memory-network / Linearly interpolated kNN over LM embedding space
DetailsInterpolates pretrained LM predictions with a kNN distribution over the LM's embedding space. Nearest neighbors drawn from any text collection enable domain adaptation without retraining.
Parameters—
Domainrag-retrievallifelong-learning
Open SourceYes
PaperView Paper
CodeRepository
iclr-2020nearest-neighbordatastorememorization

Capability Profile

Benchmark Scores

6 of 14 benchmarks
Data Transparency:6 estimated
Long-Context Retrieval
1/5
RULER
no data
NIAH
no data
LooGLE
no data
LongBench
603pEstimated
∞Bench
no data
Multi-Turn Recall
2/2
LoCoMo
64.917pEstimated
MemoryBank
617pEstimated
Cross-Session Memory
1/1
LongMemEval
68.121pEstimated
Multi-Hop QA
2/3
BABILong
78.278pEstimated
MultiHop-RAG
no data
HotpotQA
69.542pEstimated
Agent Task Memory
0/1
AgentBench-Mem
no data
Personalization
0/1
PerLTQA
no data
Factuality / Grounding
0/1
RAGAS
no data
Sources:Arena estimate — derived from capability profile, not independently verified