GitHub · by TianyuCodings
NanoJev
A 0.6B replica: parallel decisions in one forward pass, no decoding.
Geek Lite
@QingQ77
复刻 Jev 的并行决策思路，让 0.6B 小模型一次前向直接输出完整概率分布，不做输出 token 解码。 github.com/TianyuCodings/…
Sep 19, 2026 · 23 likesOpen on X
NanoJev copies Jev’s parallel decision approach in a 0.6B model. It emits the full probability distribution in a single forward pass, with no output token decoding, and ships dynamic candidates and an end-to-end training pipeline.
Open the source
