模型作者回应:Jev 证明判别式模型可统一到单一架构
一位参与新模型的作者表示,Jev 提醒了业界零样本非生成式分类模型是刚需,且这些分散在不同架构的需求可以用单一模型完成。
One of the guy who helped shape our latest model here I think the reason is much simpler, you guys highlighted that zero shot non-generative classification models are something all companies need and crave for This is something I have been trying to convince people for years and one of the reason we built ModernBERT in the first place, so I want to really thank you for that We, as most of the companies, have also strong needs for fast and powerful discriminative models. Right now they are scattered around different architecture and models, which is both a pain and inefficient Jev showed and reminded a lot of people that all of these can be done with a single model, and I am very happy to be able to try to put out there a serious and well designed model in the open that we will also be proud to us internally