This repository utilizes CUDA cuBLAS to implement a Multilayer Perceptron (MLP) for rapid network inference. The purpose of this repository is to:
- Train an MLP using Python and save the resulting weights and biases as binary files.
- Employ CUDA cuBLAS to perform inference using the previously saved weights and biases.
- Run
train.py
and we get the savedweights.bin
andbias.bin
. - Specify the paths of weights and bias in
cublas_mlp.cu
and run this script. - Finally, we can see that the python MLP and cuda MLP get the same output.