DeepSeek

Modify the Multiplication, Retain the Parameters: Associative Algebraic Layers for Transformers

This research explores replacing standard matrix multiplication in Transformer projections with a lower-rank a…

Deepseek’s Massive Fundraise Gains CATL and Tencent Backing as AI Startup Targets 2027 IPO

Deepseek seeks $12B, possibly $15B, with CATL and Tencent backing. V4-Flash benchmarks, Huawei chips, and open…

Chinese AI Industry Unites as DeepSeek Releases Open-Source Software for Huawei Ascend Chips

DeepSeek and Huawei use TileLang to optimize a 128-chip Ascend supernode, building an AI chip software ecosyst…