Unsloth 开源本地训练 Decision 模型的方法，Qwen3.5 0.8B 决策基准准确率提升至 74.3%
Unsloth 宣布用户现在可以在本地训练自己的 Decision 模型，其将 Qwen3.5 0.8B 在 3 个决策基准上的综合准确率从 20.7% 提升到 74.3%，仅需 4GB VRAM。借助开源 Unsloth repo，可把 Qwen3.8、Gemma 4 等任意 LLM 转成决策模型；具体做法是用 Unsloth 和 LoRA（r=64）微调一个 Clef head 训练一个 epoch，将下游准确率从 30–37% 提升到 78%。相关 Guide 和 Notebooks 见 unsloth.ai/docs/basics/train，仓库地址 github.com/unslothai/unsloth。
You can now train your own Decision model like Jev locally!
We increased Qwen3.5 0.8B’s aggregate accuracy from 20.7% to 74.3% across 3 decision benchmarks - on just 4GB VRAM.
Turn any LLM like Qwen3.8, Gemma 4 into decision models with our open-source Unsloth repo.
We fine-tuned with a Clef head using Unsloth and LoRA (r=64) for one epoch, increasing downstream accuracy from 30–37% to 78%.
GitHub: github.com/unslothai/unsloth
Guide and Notebooks: unsloth.ai/docs/basics/train…
来源：UnslothAI · x.com