Their point seems sound. Restated: Using pytorch [is better than] using CUDA directly (i.e., ml devs interact w/ pytorch apis not CUDA apis). And if you're usin

Their point seems sound. Restated: Using pytorch [is better than] using CUDA directly (i.e., ml devs interact w/ pytorch apis not CUDA apis). And if you're using a wrapper anyway, whether CUDA is used or some other backend is less crucial (than if directly writing to a GPGPU api) so long as the wrapper works as expected.