#128 · Primary category: Deep Learning Frameworks

open-infra-index

Production-tested AI infrastructure tools for efficient AGI development and community-driven innovation

Project last updated:05/15/25

GitHub Stars

8.1K

Forks

297

Contributors

9

License

Other

Why we included this project

DeepSeek's open-infra-index gathers the building blocks the company actually runs in production: FlashMLA attention kernels for Hopper GPUs, DeepEP for MoE all-to-all communication, DeepGEMM for FP8 matrix multiplication, the DualPipe parallelism scheme, and the 3FS file system. Each entry links to the dedicated repo and the paper behind it, so you can jump from a one-paragraph description to the code and benchmarks without wading through blog posts. The write-ups include real throughput numbers from their deployment and flag what has been battle-tested, which helps you judge whether a given kernel or library is worth evaluating for your own stack. It is an index, not a package you install, but that is exactly what makes it useful for anyone mapping out the pieces of a large-scale training or inference system.

Articles for this project

No articles for this project yet.

To suggest a topic or contribute an article, contact us.

Related projects in this category