Chinese artificial intelligence startup DeepSeek announced the commencement of its "Open Source Week" with the initial release of FlashMLA.

FlashMLA is an efficient Multi-head Latent Attention decoding kernel optimized for Hopper GPUs. It is designed to process variable-length sequences and is currently deployed in production environments, according to the company.

Last Friday, DeepSeek said it will open source five code repositories this week, describing the move as "small but sincere progress" that it will share "with full transparency."

STAFF REPORTER