量子位鹭羽
MemT 1.0 Embeds Native Memory into Transformer Architectures
Original title:RSI三块关键拼图，MemT 1.0从模型层补上「记忆」
Models68
Natively Welding Memory into Transformers
Why it's worth reading
Moving memory retention from external retrieval down into native Transformer layers presents a notable structural experiment for long-term self-improving models.
Tags
MemTTransformer模型架构记忆机制RSI自我迭代