Second Me

by Mindverse (Shang, Li, et al.)

System Card

OrganizationMindverse (Shang, Li, et al.)

Released2025-03

Architecturehybrid / Personal LPM with SFT + DPO on user data

DetailsPersonalized "Large Personal Model" acts as an intelligent memory offload: retains, organizes, and dynamically uses user-specific knowledge. Trained via automated pipeline combining supervised fine-tuning with direct preference optimization on three personal tasks.

Parameters—

Domainpersonalizationagent-memorylifelong-learning

Open SourceYes

PaperView Paper

WebsiteVisit

lpmsftdpopersonal-aiai-native-memory

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

71.931p

81.182p

Cross-Session Memory

1/1

74.944p

Multi-Hop QA

1/3

72.220p

MultiHop-RAG

no data

HotpotQA

no data

Agent Task Memory

1/1

7226p

Personalization

1/1

84.673p

Factuality / Grounding

0/1

RAGAS

no data

Sources:Second Me paper (arXiv:2503.08102); evaluated on LoCoMo: Long-Term Conversational Memory Benchmark (Snap Research, 2402)Second Me paper (arXiv:2503.08102); evaluated on LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory (Salesforce AI Research, 2410)Second Me paper (arXiv:2503.08102); evaluated on MemoryBank: Enhancing LLMs with Long-Term Memory (Sun Yat-sen University, 2305)Second Me paper (arXiv:2503.08102); evaluated on PerLTQA: A Personal Long-Term Memory Question Answering Dataset (PolyU, 2402)Second Me paper (arXiv:2503.08102); evaluated on AgentBench Memory Track (Tsinghua KEG, 2308)Second Me paper (arXiv:2503.08102); evaluated on BABILong: Testing the Limits of LLMs with Long-Context Reasoning-in-a-Haystack (AIRI, 2406)