Back to Arena

KnowAgent

by zjunlp (Zhejiang University)

System Card

Organizationzjunlp (Zhejiang University)
Released2024-03
Architectureknowledge-base / Action-knowledge base + self-learning
DetailsBuilds an "action knowledge base" converted to text for the LLM; uses knowledgeable self-learning (LoRA fine-tune on iterative trajectories) to refine plans. Works with Llama 7B/13B/70B on HotpotQA/ALFWorld.
Parameters—
Domainagent-memoryknowledge-graphlifelong-learning
Open SourceYes
PaperView Paper
CodeRepository
action-kbnaacl-2025self-learningpaper-repo

Capability Profile

Benchmark Scores

6 of 14 benchmarks
Data Transparency:1 self-reported5 estimated
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
73.739pEstimated
MemoryBank
no data
Cross-Session Memory
1/1
LongMemEval
74.642pEstimated
Multi-Hop QA
3/3
BABILong
73.535pEstimated
MultiHop-RAG
7579pEstimated
HotpotQA
48.11pSelf-Reported
Agent Task Memory
1/1
AgentBench-Mem
7226pEstimated
Personalization
0/1
PerLTQA
no data
Factuality / Grounding
0/1
RAGAS
no data
Sources:arXiv:2403.03101 Table 1 — KnowAgent-70b (Llama-2-70b-chat) F1 averaged across Easy/Medium/HardArena estimate — derived from capability profile, not independently verified