SKU AGT-4005 · Sold by External
xLAM Function-Calling 60k
Product specifications
| SKU | AGT-4005 |
|---|---|
| Data type | Agent traces |
| Volume | 60000 function-calling instances |
| Size on disk | 97.7 MB |
| Format | JSON (single file xlam_function_calling_60k.json; config 'dataset', split 'train') |
| Access model | GATED |
| Pricing | Free · gated open dataset (HF, auto-approval) |
| Quality score | — |
| License | CC BY 4.0 (Creative Commons Attribution 4.0) |
Salesforce/xlam-function-calling-60k contains 60,000 function-calling data points produced by APIGen, an automated pipeline that generates queries and the corresponding tool-call answers over 3,673 executable APIs (3,539 cleaned REST APIs sourced from ToolBench/RapidAPI plus 134 Python functions) spanning 21 categories. Each record has three fields: a natural-language query, a tools array listing the available functions (name, description, typed parameters), and an answers array giving the reference function call(s) with their arguments. Every instance is verified through APIGen's three hierarchical stages: format checking, actual function execution against real backends, and LLM-based semantic checking, and the released set was additionally human-evaluated on 600 sampled points. Queries span four styles: simple, multiple, parallel, and parallel-multiple function calls. The data was generated with DeepSeek-V2-Chat (33,659 entries) and Mixtral-8x22B-Instruct (the remaining entries). Access is gated on Hugging Face (auto-approval requiring login plus name/country/affiliation and acknowledgement of the license and APIGen citation).