tl;dr;

No CUDA, no problem. You can still train and run ML models on your Mac with CoreML. You can even convert your PyTorch and TensorFlow models to CoreML models.

Apple Machine Learning

https://coremltools.readme.io/docs/unified-conversion-api

And you can run predictions
Images need to be PIL images, not numpy arrays.

https://coremltools.readme.io/docs/model-prediction