AI Roundup: DeepSeek’s Big Moves, Claude 3.7, and the Latest Breakthroughs
Deep Papers
00:00
Intro
This chapter explores DeepSeek's Open Source Week, showcasing the debut release of FlashMLA, a super-efficient decoding kernel for NVIDIA's Hopper GPUs. It highlights the impressive performance improvements over existing technologies like Flash Attention and discusses the implications for developers in the realm of language models.
Transcript
Play full episode
