---
license: mit
language:
- en
tags:
- reasoning,
- chat,
- instruct,
- base,
pretty_name: Dual Mode Reasoning Example
size_categories:
- n<1K
---

![image/png](https://cdn-uploads.huggingface.co/production/uploads/63f2b7bcbe95ed4c9a9e7669/xYf0GGeL4sxo4NBA1WiGE.png)

This is an example of a dataset that can be used to finetune a base llm to have dual thinking modes by adding flags to the system prompt. Here is an example of a chat template (https://huggingface.co/AIGym/smollm3-base-3b/blob/main/chat_template.jinja).


If you are interested in pretraining a base model checkout this example (https://github.com/vibe-coding-projects/build-nanogpt)