<ul><li>DeepSeek has released FlashMLA, a decoding kernel for Hopper GPUs, to improve computational efficiency for AI applications.</li><li>The kernel supports variable-length sequences and achieves high performance speeds of up to 3000 GB/s for memory tasks and 580 TFLOPS for computing.</li><li>FlashMLA is available on GitHub for exploration and use, and is expected to impact sectors like cryptocurrency trading algorithms.</li><li>DeepSeek recently announced the launch of five open-source repositories and has a collection of 14 open-source models and repositories on Hugging Face.</li></ul>

DeepSeek Launches FlashMLA, an MLA Decoding Kernel for Hopper GPUs

Discover more