Software that turns an AI model into fast code for a specific device.
It is like a tailor cutting one suit design for each body. The same suit fits a small kid or a tall basketball player.
It helps AI models run faster on phones and GPUs. It also helps them use less power and memory.
ONNX
ONNX gives the compiler a shared way to read models.
CUDA
The AI Compiler can map model work to CUDA.
Inference engine
The Inference engine uses compiled results to run models faster.
GPU
The AI Compiler plans work and memory for a GPU.