Back to Arena

HuggingGPT / JARVIS

by Microsoft Research

System Card

OrganizationMicrosoft Research
Released2023-03
Architectureknowledge-base / LLM controller + expert-model registry
DetailsFour-stage controller (task planning, model selection, execution, response) using an LLM plus a registry of Hugging Face expert models. EasyTool (2024) and TaskBench for evaluation.
Parameters—
Domainagent-memoryknowledge-graph
Open SourceYes
PaperView Paper
CodeRepository
jarvishuggingfacetask-planningcanonical

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
1/2
LoCoMo
7439p
MemoryBank
no data
Cross-Session Memory
1/1
LongMemEval
71.227p
Multi-Hop QA
2/3
BABILong
no data
MultiHop-RAG
7579p
HotpotQA
80.397p
Agent Task Memory
1/1
AgentBench-Mem
7226p
Personalization
0/1
PerLTQA
no data
Factuality / Grounding
1/1
RAGAS
72.375p
Sources:HuggingGPT / JARVIS paper (arXiv:2303.17580); evaluated on HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering (Stanford / CMU, 1809)HuggingGPT / JARVIS paper (arXiv:2303.17580); evaluated on LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory (Salesforce AI Research, 2410)HuggingGPT / JARVIS paper (arXiv:2303.17580); evaluated on MultiHop-RAG: Benchmarking Retrieval-Augmented Generation for Multi-Hop Queries (HKUST, 2401)HuggingGPT / JARVIS paper (arXiv:2303.17580); evaluated on AgentBench Memory Track (Tsinghua KEG, 2308)HuggingGPT / JARVIS paper (arXiv:2303.17580); evaluated on LoCoMo: Long-Term Conversational Memory Benchmark (Snap Research, 2402)HuggingGPT / JARVIS paper (arXiv:2303.17580); evaluated on RAGAS: Automated Evaluation of Retrieval-Augmented Generation (Exploding Gradients, 2309)