SKU AGT-4006 · Sold by External
Glaive Function Calling v2
Product specifications
| SKU | AGT-4006 |
|---|---|
| Data type | Agent traces |
| Volume | 112960 chat samples (multi-turn function-calling conversations) |
| Size on disk | ~271 MB (glaive-function-calling-v2.json, 271,190,065 bytes download_size; 254,485,835 bytes parsed dataset_size) |
| Format | JSON (single glaive-function-calling-v2.json file; two string fields: system, chat) |
| Access model | PUBLIC LICENSE |
| Pricing | Free · open dataset (HF) |
| Quality score | — |
| License | Apache License 2.0 |
Glaive Function Calling v2 is a synthetic instruction-tuning dataset for teaching LLMs to invoke tools/functions. Each record is a single-turn or multi-turn conversation: a `system` field lists one or more callable functions as JSON schemas (name, description, parameters with types and required fields), and a `chat` field contains the full USER / ASSISTANT / FUNCTION RESPONSE dialogue serialized as one string, where the assistant emits tool calls as `<functioncall> {json}` and turns are terminated by `<|endoftext|>`. The corpus includes both positive cases (the assistant selects and calls the right function with correctly-typed arguments) and negative cases (the assistant declines when no available tool matches the request). The dataset has exactly 112,960 examples (~254 MB uncompressed, distributed as a single 271 MB JSON file). It is public and ungated, licensed Apache-2.0, authored by Glaive AI; the Hugging Face card is metadata-only (no prose description, citation, or Data Fields section beyond the two `system`/`chat` string features).