1 paper
Zhiheng Zhang, Mujie Xu, Feiyu Sun +1
LLMs generate tool calls token by token, even though the function choice and argument values can often be predicted in parallel from the request and tool schema. ToolSpec reduces t…