ToolGrad: Efficient tool-use dataset generation with textual "gradients"
Google Research Blog
Read full post
Researchers introduced ToolGrad, a new method to efficiently generate datasets for training large language models (LLMs) on tool-use tasks by first creating tool-use chains and then annotating user prompts. This approach produces more complex data at lower cost and improves LLM performance, matching state-of-the-art proprietary models on unseen tools.
