ToolGrad-500: Synthetic Tool-Use Data for Parallel Function Calling

ToolGrad-500 is a synthetic tool-use dataset generated by the ToolGrad framework. It targets single-turn parallel function calling scenarios, optimized using textual gradients. The dataset was created by zhongyi-zhou and updated on June 1, 2026.

Use Cases

• Fine-tuning language models for tool selection based on textual gradients.
• Training models on parallel function calling scenarios based on the single-turn conversation format.
• Benchmarking model performance on synthetic tool-use tasks.
• Researching methods for optimizing function calling in conversational agents.

Strengths

• Dataset is generated using the ToolGrad framework, a method published at ACL 2026 Findings.
• Specifically targets single-turn parallel function calling, a defined and relevant task.

Limitations

• Dataset size, row count, and file formats are unknown, limiting suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• Data is synthetic, which may limit its realism or applicability to real-world tool-use scenarios.

Provenance

Source

zhongyi-zhou via Hugging Face.

Collection Method

Synthetically generated using the ToolGrad framework leveraging textual gradients.

Freshness

Last updated 2026-06-01 03:13:59.