Back to Arena

Memoripy

by caspianmoon

System Card

Organizationcaspianmoon
Released2024-11
Architecturehybrid / Short/long-term with Ebbinghaus decay + concept graph
DetailsShort-term + long-term dual store with automatic promotion based on relevance, graph-based concept associations with spreading activation, hierarchical semantic clustering, and time-decay that reinforces frequently accessed memories.
Parameters—
Domainagent-memorypersonalization
Open SourceYes
CodeRepository
ebbinghaus-decayconcept-graphshort-long-termpython

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
75.148p
MemoryBank
74.136p
Cross-Session Memory
1/1
LongMemEval
73.636p
Multi-Hop QA
2/3
BABILong
no data
MultiHop-RAG
70.338p
HotpotQA
70.145p
Agent Task Memory
1/1
AgentBench-Mem
7226p
Personalization
0/1
PerLTQA
no data
Factuality / Grounding
0/1
RAGAS
no data
Sources:Memoripy (caspianmoon/memoripy); evaluated on LoCoMo: Long-Term Conversational Memory Benchmark (Snap Research, 2402)Memoripy (caspianmoon/memoripy); evaluated on LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory (Salesforce AI Research, 2410)Memoripy (caspianmoon/memoripy); evaluated on AgentBench Memory Track (Tsinghua KEG, 2308)Memoripy (caspianmoon/memoripy); evaluated on HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering (Stanford / CMU, 1809)Memoripy (caspianmoon/memoripy); evaluated on MemoryBank: Enhancing LLMs with Long-Term Memory (Sun Yat-sen University, 2305)Memoripy (caspianmoon/memoripy); evaluated on MultiHop-RAG: Benchmarking Retrieval-Augmented Generation for Multi-Hop Queries (HKUST, 2401)