Unsloth open-sources tutorial for training local decision models
What happened
On October 8, Unsloth released a tutorial and open-source repo showing how to fine-tune large language models such as Qwen3.8 and Gemma 4 into decision models that output probabilities over options. It also reports training results for Qwen3.5 0.8B: combined accuracy on three decision benchmarks rose from 20.7% to 74.3%. The local training described in the tutorial needs only 4GB of VRAM. So far the release includes the training tutorial and the open-source repo.
Written by AI from the coverage · updated 1 hour ago
Coverage
Follow the reports to see the story from different sides.
- AI HOT · Tips & opinionsUnsloth 开源教程：本地训练 Qwen3.5 0.8B 决策模型，准确率从 20.7% 提升至 74.3%
Unsloth 发布教程与开源仓库，可将 Qwen3.8、Gemma 4 等 LLM 微调为输出选项概率的决策模型。Qwen3.5 0.8B 在 3 个决策基准上的合计准确率从 20.7% 提升至 74.3%，仅需 4GB 显存即可本地训练。
Heat over time
Not enough continuous observations to draw a trend yet.