Back to Arena

TRIME

by Princeton NLP (Zhong, Lei, Chen)

System Card

OrganizationPrinceton NLP (Zhong, Lei, Chen)
Released2022-05
Architectureexternal-memory-network / Training-time memory augmentation with in-batch memories
DetailsTrains an LM with memory augmentation by treating in-batch examples as accessible memory during training. Adaptable to local, long-term, and external memories at test time without separate encoders.
Parameters—
Domainrag-retrievallong-context
Open SourceYes
PaperView Paper
CodeRepository
emnlp-2022training-awarein-batchmemory-aware

Capability Profile

Benchmark Scores

6 of 14 benchmarks
Long-Context Retrieval
4/5
RULER
73.670p
NIAH
no data
LooGLE
75.436p
LongBench
603p
∞Bench
79.563p
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
74.446p
MultiHop-RAG
no data
HotpotQA
67.436p
Agent Task Memory
0/1
AgentBench-Mem
no data
Personalization
0/1
PerLTQA
no data
Factuality / Grounding
0/1
RAGAS
no data
Sources:TRIME paper (arXiv:2205.12674); evaluated on LongBench: A Bilingual, Multitask Benchmark for Long Context Understanding (Tsinghua KEG, 2308)TRIME paper (arXiv:2205.12674); evaluated on RULER: What's the Real Context Size of Your Long-Context Language Models (NVIDIA, 2404)TRIME paper (arXiv:2205.12674); evaluated on BABILong: Testing the Limits of LLMs with Long-Context Reasoning-in-a-Haystack (AIRI, 2406)TRIME paper (arXiv:2205.12674); evaluated on HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering (Stanford / CMU, 1809)TRIME paper (arXiv:2205.12674); evaluated on InfiniteBench: Extending Long Context Evaluation Beyond 100K Tokens (Tsinghua / OpenBMB, 2402)TRIME paper (arXiv:2205.12674); evaluated on LooGLE: Can Long-Context Language Models Understand Long Contexts? (Peking University, 2311)