Search
DeepSeek Unveils New Research Paper Authored by Liang Wenfeng, Makes Related Memory Module Engram Open-Source
3 week ago /
Read about 0 minute
Author：小编
On the evening of the 12th, DeepSeek, in partnership with Peking University, published a new research paper named "Conditional Memory with Scalable Lookup: A New Dimension of Sparsity for Large Language Models", with Liang Wenfeng as one of the co-authors. In this paper, the innovative concept of conditional memory is introduced, which is achieved by integrating a scalable lookup memory structure. This breakthrough significantly boosts the model's capabilities in knowledge retrieval, reasoning, coding, and mathematical problem-solving, all while keeping the number of parameters and computational resources constant. Moreover, DeepSeek has decided to make the relevant memory module, Engram, available as open-source.
-
C114 Communication Network -
Communication Home
7 X 24 Track global technological trends
