DeepSeek V4.1 Flash支持n-gram块编辑
精选理由
DeepSeek让模型能像编辑文档一样添加新知识，几MB就能更新事实，比传统微调便宜多了。
DeepSeek V4.1 Flash模型在LongCat测试中实现了n-gram块编辑功能。用户可通过添加几MB大小的覆盖层引入新事实。该技术支持基于三元组[主体、关系、客体]的持续学习。
原文 · Teortaxes
n-gram blocks in models like DeepSeek V4.1 Flash (experiments are on LongCat) can be edited, with a cheaply derived overlay a few MBs in size, introducing new declarative facts. Primitive (only [subject, relation, object] triples) but continual learning per-user is possible. https://t.co/taEYpeOz8Q
