返回探索

TypeSafe:不花十亿预训练,选择拼接现有模型

TypeSafe的Diogo Almeida表示,即使有十亿美元也不会从头预训练基础模型,而是通过拆分、重组现有模型来构建Jev,强调避免烧钱训练模型。

Give TypeSafe’s Diogo Almeida a billion dollars and he still wouldn’t pre-train. His bet: slice, dice, and Frankenstein existing stacks — anything except burning the check on a scratch foundation model. @swyx @labenz Latent Space — Why We Made Jev — Diogo Almeida, TypeSafe

· 0 次赞在 X 打开