Glaive Function Calling v2 converted into the ms-swift Agent format for supervised fine-tuning. The dataset contains approximately 109,000 JSONL entries for training AI agents to use tools. It was created by author hhzhou and last updated on June 21, 2026.
Use Cases
- Supervised fine-tuning (SFT) of language models for tool invocation based on the described agent format.
- Low-rank adaptation (LoRA) training for specialized function-calling tasks.
- Benchmarking and smoke testing agent performance on tool-use scenarios.
- Training AI assistants to parse and execute structured function calls.
Strengths
- Provides a substantial training set of approximately 109,000 examples.
- Includes curated subsets (10k and 1k entries) for rapid prototyping and evaluation.
- Data is formatted specifically for the ms-swift and Hermes Agent templates.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count for the primary file is an approximation (~109k), and other size metrics are unknown.
- Description metadata is limited; actual data quality requires manual inspection after download.
Provenance
- Source
- Converted from Glaive Function Calling v2.
- Collection Method
- Converted using a script (data_process/glaive_to_ms_swift.py).
- Time Range
- null
- Freshness
- Last updated 2026-06-21 03:21:31; freshness should be verified.
- Geography
- null