GPU 算子 / 通信

3 条

Attention 算子与集合通信库。

#仓库Stars
1deepseek-ai/DeepEP
DeepEP: an efficient expert-parallel communication library
10k
2linkedin/Liger-Kernel
Efficient Triton Kernels for LLM Training
6.7k
3NVIDIA/nccl
Optimized primitives for collective multi-GPU communication
5.2k