xLam Function Calling 60k
This dataset contains 60,000 sample text queries with associated functional API calls, automatically verified by execution and semantic validation, covering 21 API categories. It is designed for the development of function-calling models that can interpret and execute functions from natural text.
60,000 JSON examples, structured format: request, tools, parameters, responses
CC-BY 4.0
Description
The dataset xLam Function Calling 60k includes 60,000 sample natural language requests, combined with functional API calls that are validated through three steps: format control, actual execution, and semantic verification. The data covers 21 diverse API categories and is generated by two automatic pipelines.
What is this dataset for?
- Train models that can call API functions from natural language instructions
- Evaluate the accuracy of the models in understanding the parameters and correct execution
- Benchmark function-calling agents in various contexts
Can it be enriched or improved?
This dataset can be enriched by adding new types of APIs, integrating execution logs, or creating multilingual versions. Finer annotation of minor errors detected could also improve robustness.
🔎 In summary
🧠 Recommended for
- NLP researchers
- API assistant developers
- Function-calling ML engineers
🔧 Compatible tools
- Hugging Face Datasets
- LangChain
- OpenAI Function Calling
- FastAPI
💡 Tip
Use semantic validation as a filter to detect and correct errors in your own data.
Frequently Asked Questions
What is a function-calling agent in the context of this dataset?
A function-calling agent interprets instructions in natural language to automatically execute calls to API functions with the appropriate parameters.
Does the dataset contain examples of errors or incorrect data?
Yes, around 5% of the examples have small errors detected during validation, such as inaccurate arguments, but overall the quality is greater than 95%.
Can this dataset be used for commercial applications?
Yes, the CC-BY 4.0 license allows commercial use as long as attribution is mentioned.
