# Prism DualModel Reasoning

This dataset contains synthetic solution traces. Exactly 60% of the released traces were generated with [Qwen/Qwen2.5-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-7B-Instruct), while exactly 40% were generated with [microsoft/Phi-3-mini-4k-instruct](https://huggingface.co/microsoft/Phi-3-mini-4k-instruct).

The model shares refer to final released records after filtering. Neither model revised the other model's traces.

The prompts were sampled from the companion Prism MathBlend page, but no source answers or solution text were copied. [mistralai/Mistral-7B-Instruct-v0.3](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.3) was used only as an evaluation judge and contributed no released trace.

Project utilities are hosted in the [Prism-LicenseDesk repository](https://github.com/toolathlon-jd/Prism-LicenseDesk), and companion releases are listed under [toolathlonEval](https://huggingface.co/toolathlonEval).