Back to Arena

Synapse

by NTU / Salesforce (Zheng et al.)

System Card

OrganizationNTU / Salesforce (Zheng et al.)
Released2023-06
Architectureepisodic-buffer / Trajectory-as-exemplar prompting with exemplar memory
DetailsState abstraction compresses raw HTML into task-relevant observations; trajectory-as-exemplar prompting uses full abstracted action sequences retrieved from an exemplar memory via similarity search.
Parameters—
Domainagent-memoryepisodic-session
Open SourceYes
PaperView Paper
WebsiteVisit
CodeRepository
iclr-2024web-agentsminiwobtrajectory

Capability Profile

Benchmark Scores

6 of 14 benchmarks
Long-Context Retrieval
0/5
RULER
no data
NIAH
no data
LooGLE
no data
LongBench
no data
∞Bench
no data
Multi-Turn Recall
2/2
LoCoMo
71.830p
MemoryBank
7867p
Cross-Session Memory
1/1
LongMemEval
7439p
Multi-Hop QA
2/3
BABILong
no data
MultiHop-RAG
7250p
HotpotQA
67.436p
Agent Task Memory
1/1
AgentBench-Mem
7226p
Personalization
0/1
PerLTQA
no data
Factuality / Grounding
0/1
RAGAS
no data
Sources:Synapse paper (arXiv:2306.07863); evaluated on AgentBench Memory Track (Tsinghua KEG, 2308)Synapse paper (arXiv:2306.07863); evaluated on LoCoMo: Long-Term Conversational Memory Benchmark (Snap Research, 2402)Synapse paper (arXiv:2306.07863); evaluated on LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory (Salesforce AI Research, 2410)Synapse paper (arXiv:2306.07863); evaluated on HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering (Stanford / CMU, 1809)Synapse paper (arXiv:2306.07863); evaluated on MemoryBank: Enhancing LLMs with Long-Term Memory (Sun Yat-sen University, 2305)Synapse paper (arXiv:2306.07863); evaluated on MultiHop-RAG: Benchmarking Retrieval-Augmented Generation for Multi-Hop Queries (HKUST, 2401)