Upload PyTorch quantized model (pure PyTorch, no custom CUDA kernels)

initial commit