A high-throughput and memory-efficient inference and serving engine for LLMs

VideoNSA: Native Sparse Attention Scales Video Understanding

An intuitive and low-overhead instrumentation tool for Python

A Quirky Assortment of CuTe Kernels

Domain-specific language designed to streamline the development of high-performance GPU/CPU/Accelerators kernels

Efficient Triton Kernels for LLM Training

Ongoing research training transformer models at scale

🚀 Efficient implementations of state-of-the-art linear attention models

Development repository for the Triton language and compiler

DeepGEMM: clean and efficient FP8 GEMM kernels with fine-grained scaling

qwen-nsa

FlashMLA: Efficient Multi-head Latent Attention Kernels

OpenSeek aims to unite the global open source community to drive collaborative innovation in algorithms, data and systems to develop next-generation models.