Instruction Datasets for Qwen3 Tool Calling and Reasoning

A collection of five datasets converted and tokenized for the Qwen3 large language model. The datasets focus on sequential tool use, single-step reasoning, multi-turn reasoning, and function calling with chain-of-thought. The collection was prepared by author jtl11 and was last updated in August 2025.

Use Cases

• Fine-tuning models on sequential tool use tasks using the 'toolace_sequential_tool_use_reasoning' dataset.
• Training models for single-turn tool invocation and reasoning with the 'tool-calls-single-reasoning' dataset.
• Developing conversational agents capable of multi-turn tool use via the 'tool-use-multiturn-reasoning' dataset.
• Improving function calling with chain-of-thought reasoning using the 'function-calling-v0.2-with-r1-cot' dataset.
• Enhancing general function calling and reasoning capabilities with the 'function-calling-reasoning-v1' dataset.

Strengths

• Comprises five distinct datasets focused on specialized LLM tasks.
• Specifically processed and tokenized for compatibility with the Qwen3 model architecture.

Limitations

• No information on dataset size, row count, or sample data is provided.
• Original column structures and data formats for the constituent datasets are unknown.

Provenance

Source

Aggregated from five datasets on the Hugging Face platform by user interstellarninja, AymanTarig, and zzzch.

Collection Method

Converted and tokenized from the original source datasets.

Freshness

Last updated on 2025-08-01.