Club&Lab for PaddlePaddle contributors

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

A Quirky Assortment of CuTe Kernels

MoonEP: A Perfectly Balanced Expert Parallelism Library via Dynamic Redundant Experts

cuDNN Frontend is NVIDIA's modern, open-source entry point to the cuDNN library and a growing collection of high-performance open-source kernels.

An extended sonic-moe implementation, with FP8 support fully developed by agents

MCore-Bridge: Providing Megatron-Core model definitions for state-of-the-art large models and making Megatron training as simple as Transformers — with support for 300+ large language models (Qwen3-Next, GLM-5.2, Deepseek-V4, MiniMax-2.7, ...) and 200+ multimodal large models (Qwen3.5, Qwen3-Omni, Gemma4, ...).