A JEV-style decision model reportedly improved from 37% to 65% accuracy in 10 minutes of local training

A user says they trained a JEV-style decision model locally with Qwen3.5 0.8B as its base. They report that accuracy rose from 37% to 65% in 60 steps, or 10 minutes of training, using 4GB of VRAM. They also share a video walkthrough covering Unsloth Studio setup, LoRA training, testing and inference.