facebookincubator/AITemplate

AITemplate is a Python framework which renders neural network into high performance CUDA/HIP C++ code. Specialized for FP16 TensorCore (NVIDIA GPU) and MatrixCore (AMD GPU) inference.

177 best points · 65 comments · 3 Hacker News posts · 4,735 GitHub stars · View on GitHub

Hacker News discussions

• Render a neural network into CUDA/HIP code (177 points, 65 comments, by fzliu, 2023-06-02)
• PyTorch models to optimised C++ with 2x speedup (3 points, 0 comments, by nl, 2022-10-04)
• AITemplate for Stable Diffusion (1 points, 0 comments, by hochmartinez, 2022-10-04)