HyperAI

External Tensor Functions

Although TVM supports transparent code generation, sometimes you need to merge handwritten code into the pipeline, such as using cuDNN for some convolution kernels and defining the remaining stages.