Trending storyDeveloping
Jeff: 0.8B decision models trained at home, ~30 ms inference
1 report1 sourceupdated 21 hours ago
What happened
Summary
Jeff 是一组从 Qwen3.5 和 Gemma 4 微调出来的 0.8B 参数小模型，专门做零样本分类——也就是不给例子直接判断类别。亮点是作者说在家用普通显卡上就能训练，推理一次只要 30 毫秒左右，延迟很低，适合塞进实时决策流程。模型兼容 Jev 格式，GitHub 上放了代码和权重。不过正文没披露用了多大训练集、在什么基准上测过，所以实际效果...
Coverage
Follow the reports to see the story from different sides.
Sep 29
- Hacker News front pageJeff: 0.8B decision models trained at home, ~30 ms inference
Jeff is a set of 0.8B parameter models fine-tuned from Qwen3.5 and Gemma 4 for zero-shot classification. Trained on consumer hardware at home, it runs inference in ~30 ms and is Jev-compatible. The post doesn't disclose dataset size or benchmarks, but the GitHub repo includes code and weights. For teams needing lightweight decision pipelines, the latency and size are practical.
Heat over time
Not enough continuous observations to draw a trend yet.