Research map / Large language models / AI and society
HuggingGPT: Solving AI Tasks with ChatGPT and its Friends in Hugging Face
Yongliang Shen, Kaitao Song, Xu Tan, Dongsheng Li, Weiming Lu, Yueting Zhuang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use GPT-3.5 instead of Vicuna-7b as controller | HuggingGPT · single task planning | accuracy: 23.86 -> 52.62 | Improved |
| use GPT-3.5 instead of Vicuna-7b as controller | HuggingGPT · sequential task planning | F1: 22.89 -> 51.92 | Improved |
| use GPT-3.5 instead of Vicuna-7b as controller | HuggingGPT · graph task planning | GPT-4 Score: 19.17 -> 50.48 | Improved |
| use GPT-4 instead of GPT-3.5 as controller | HuggingGPT · human-annotated sequential planning | accuracy: 18.18 -> 41.36 | Improved |
| use GPT-4 instead of GPT-3.5 as controller | HuggingGPT · human-annotated graph planning | F1: 16.45 -> 49.28 | Improved |
| use GPT-3.5 instead of Vicuna-13b as controller | HuggingGPT · human evaluation, 130 requests | success rate (%): 15.64 -> 63.08 | Improved |
| increase demonstration variety from 2 to 10 task types (ablation) | HuggingGPT with GPT-3.5 · single task planning | accuracy: 43.31 -> 52.83 | Improved |
| increase demonstration variety from 2 to 10 task types (ablation) | HuggingGPT with GPT-4 · single task planning | accuracy: 65.59 -> 67.52 | Improved |
| increase demonstration variety from 2 to 10 task types (ablation) | HuggingGPT with GPT-3.5 · graph task planning | F1: 43.42 -> 64.24 | Improved |
| increase demonstration variety from 2 to 10 task types (ablation) | HuggingGPT with GPT-4 · sequential task planning | edit distance (%): 47.17 -> 39.32 | Improved |
| increase demonstration variety from 2 to 6 task types (ablation) | HuggingGPT with GPT-3.5 · sequential task planning | F1: 32.15 -> 43.19 | Improved |
| increase number of demonstrations beyond 4 (ablation) | HuggingGPT · task planning | accuracy | No clear effect |
Closest papers by meaning
- NExT-GPT: Any-to-Any Multimodal LLM (ICML 2024 · similarity 0.86)
- AudioGPT: Understanding and Generating Speech, Music, Sound, and Talking Head (AAAI 2024 · similarity 0.85)
- GPTSwarm: Language Agents as Optimizable Graphs (ICML 2024 · similarity 0.85)
- ShortcutsBench: A Large-Scale Real-world Benchmark for API-based Agents (ICLR 2025 · similarity 0.84)
- On Path to Multimodal Generalist: General-Level and General-Bench (ICML 2025 · similarity 0.84)
- PortLLM: Personalizing Evolving Large Language Models with Training-Free and Portable Model Patches (ICLR 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
