Back to Arena

Lagent

by InternLM (Shanghai AI Lab)

System Card

OrganizationInternLM (Shanghai AI Lab)
Released2023-08
Architectureagentic-workflow / PyTorch-style agent with get_memory()/state_dict()
DetailsLightweight, PyTorch-inspired agent framework. Messages auto-accumulate in agent memory during forward passes; accessible via `get_memory()` / `state_dict()`. Async variants, custom aggregators for few-shot prompting.
Parameters—
Domainagent-memory
Open SourceYes
CodeRepository
pytorch-likeinternlmlightweightstate-dict

Capability Profile

Benchmark Scores

5 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
1/2
LoCoMo
74.643p
MemoryBank
no data
Cross-Session Memory
1/1
LongMemEval
80.482p
Multi-Hop QA
2/3
BABILong
no data
MultiHop-RAG
74.373p
HotpotQA
72.859p
Agent Task Memory
1/1
AgentBench-Mem
7226p
Personalization
0/1
PerLTQA
no data
Factuality / Grounding
0/1
RAGAS
no data
Sources:Lagent (InternLM/lagent); evaluated on AgentBench Memory Track (Tsinghua KEG, 2308)Lagent (InternLM/lagent); evaluated on HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering (Stanford / CMU, 1809)Lagent (InternLM/lagent); evaluated on LoCoMo: Long-Term Conversational Memory Benchmark (Snap Research, 2402)Lagent (InternLM/lagent); evaluated on LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory (Salesforce AI Research, 2410)Lagent (InternLM/lagent); evaluated on MultiHop-RAG: Benchmarking Retrieval-Augmented Generation for Multi-Hop Queries (HKUST, 2401)