Laya Just Dropped: A 421M Model That Decides in 33ms
中文摘要
Laya 是一个 4.21 亿参数的模型，决策仅需 33 毫秒。其核心在于不生成 token，这既是其独特机制也是其局限。
English Summary
Laya is a 421M parameter model that decides in 33ms. It uniquely avoids generating tokens, a method that serves as both its core trick and its catch.
原文节选
It never writes a single token. That’s the whole trick — and the catch. Continue reading on Generative AI »