大模型NeurIPS 20252025
Toolformer 2: Language Models as Tool Users
Timo Schick, Jane Dwivedi-Yu, Roberto Dessi, Roberto NavigliMeta AI, Sapienza University
摘要
We present Toolformer 2, which enables language models to autonomously use external tools including calculators, search engines, and APIs. Toolformer 2 learns to decide when and how to use tools through self-supervised training, significantly expanding the capabilities of language models for practical applications.
tool uselanguage modelself-supervisedAPI integrationpractical application
技术细节
数据集
Toolformer DatasetAPI-BankToolBench
模型骨架
LLaMA-2 13B + Tool Calling Module
编码器
Transformer Encoder + Tool Description Encoder
解码器
Tool-Augmented Decoder (32 layers)