by Salesforce
xLAM Function-Calling 60k is Salesforce AI Research's dataset of 60,000 verified examples of turning plain-language requests into API calls. It is for teams training or evaluating function-calling language models and agents.
DatasetFree
Step 1: Sign in to Hugging Face and accept the dataset's access conditions on its page.
Step 2: Install the datasets library with pip install datasets and authenticate with your Hugging Face token.
Step 3: Load it with load_dataset("Salesforce/xlam-function-calling-60k").
Step 4: Parse the query, tools and answers fields with json.loads, since they are stored as strings.
Step 5: Use the examples to fine-tune or evaluate a model's function-calling output, and cite APIGen.
CC-BY-4.0. You must accept the conditions on Hugging Face, including citing APIGen, to access the files.
Each example went through format checking, real function execution and semantic verification. A human review of 600 samples found a correct rate above 95%.
Two open-weights language models generated the entries; the dataset card names them, and each example was verified afterward.
JSON. Each entry has a query, a list of tools with parameters, and the expected answers as tool names with arguments.