Unsloth offers free notebook to train Qwen3.5-4B decision models on 8GB VRAM

Unsloth AI says it has released a free notebook that lets users train Qwen3.5-4B to output decisions rather than free text, running locally on about 8GB of VRAM. According to the company's announcement on X, the walkthrough covers preparing training data made of a state, questions, and gold answers, then training the model and serving it locally.

The claim rests on Unsloth's own announcement; the report includes no independent testing of the model's decision quality or the VRAM requirement in practice.