Documentation on adapter train is lacking any details related to training on dataset with tool calling. And page about tool calling itself only explain how to use it from Swift without any internal details useful in training.
Question is how schema should looks like for including tool calling in dataset?