DeepSeek AI Ends Open Source Week with Fire-Flyer File System (3FS) for AI Data Access
It’s the last day of DeepSeek’s Open Source Week, and they’re finishing with a bang! The Fire-Flyer File System (3FS) is their final release – a game-changing distributed file system. The DeepSeek 3FS fixes a big problem in AI development: getting data in and out fast enough. As AI models get bigger and datasets grow […]
DeepSeek Open Source Week Day 4, Optimized Parallelism Strategies with DualPipe, EPLB and Profile Data
DeepSeek AI just released some amazing open-source tools on the 4th day of their Open Source Week. They’re called Optimized Parallelism Strategies and include cool innovations like DualPipe, Expert Parallelism Load Balancer (EPLB), and Profile Data for their DeepSeek V3 and R1 models. These tools help make AI run faster and more efficiently, which is […]
DeepSeek AI Drops DeepGEMM, An FP8 GEMM Library That Powers V3 and R1 AI Models
It is Day 3 of DeepSeek Open Source Week, and the company released its 3rd bomb: DeepGEMM. Following the release of FlashMLA on the first day and DeepEP on the second day, DeepSeek is making powerful open-source contributions to the AI community. DeepGEMM stands out as an efficient FP8 library to make AI computations faster […]
DeepSeek AI Releases DeepEP, A High-Performance MoE Library on Day 2 of Open Source Week
DeepSeek AI just dropped ‘DeepEP‘ on the second day of their Open Source Week event. DeepEP is a special tool that helps AI models work better and faster, especially Mixture-of-Experts (MoE) models. This announcement comes right after DeepSeek’s first-day release of FlashMLA, an efficient MLA decoding kernel designed for Hopper GPUs. In this world of […]
DeepSeek AI Introduces FlashMLA, The Powerhouse Kernel for Large-Scale AI Models
DeepSeek, a Chinese AI lab, recently announced it would open-source parts of its technology as part of an event called “Open Source Week.” The company shared plans to release five code repositories, all of which have already been tested and used in real-world AI applications. Today, the event kicked off with the launch of FlashMLA, […]