AgentBank: Towards Generalized LLM Agents via Fine-Tuning on 50000+ Interaction Trajectories
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2024
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866909344243646464 |
|---|---|
| author | Song, Yifan Xiong, Weimin Zhao, Xiutian Zhu, Dawei Wu, Wenhao Wang, Ke Li, Cheng Peng, Wei Li, Sujian |
| author_facet | Song, Yifan Xiong, Weimin Zhao, Xiutian Zhu, Dawei Wu, Wenhao Wang, Ke Li, Cheng Peng, Wei Li, Sujian |
| contents | Fine-tuning on agent-environment interaction trajectory data holds significant promise for surfacing generalized agent capabilities in open-source large language models (LLMs). In this work, we introduce AgentBank, by far the largest trajectory tuning data collection featuring more than 50k diverse high-quality interaction trajectories which comprises 16 tasks covering five distinct agent skill dimensions. Leveraging a novel annotation pipeline, we are able to scale the annotated trajectories and generate a trajectory dataset with minimized difficulty bias. Furthermore, we fine-tune LLMs on AgentBank to get a series of agent models, Samoyed. Our comparative experiments demonstrate the effectiveness of scaling the interaction trajectory data to acquire generalized agent capabilities. Additional studies also reveal some key observations regarding trajectory tuning and agent skill generalization. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2410_07706 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | AgentBank: Towards Generalized LLM Agents via Fine-Tuning on 50000+ Interaction Trajectories Song, Yifan Xiong, Weimin Zhao, Xiutian Zhu, Dawei Wu, Wenhao Wang, Ke Li, Cheng Peng, Wei Li, Sujian Computation and Language Artificial Intelligence Fine-tuning on agent-environment interaction trajectory data holds significant promise for surfacing generalized agent capabilities in open-source large language models (LLMs). In this work, we introduce AgentBank, by far the largest trajectory tuning data collection featuring more than 50k diverse high-quality interaction trajectories which comprises 16 tasks covering five distinct agent skill dimensions. Leveraging a novel annotation pipeline, we are able to scale the annotated trajectories and generate a trajectory dataset with minimized difficulty bias. Furthermore, we fine-tune LLMs on AgentBank to get a series of agent models, Samoyed. Our comparative experiments demonstrate the effectiveness of scaling the interaction trajectory data to acquire generalized agent capabilities. Additional studies also reveal some key observations regarding trajectory tuning and agent skill generalization. |
| title | AgentBank: Towards Generalized LLM Agents via Fine-Tuning on 50000+ Interaction Trajectories |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2410.07706 |