I use an Air for ML. It's perfect if you work in the cloud. Locally it's too much hassle. Not just because of the limited power, but because of lack of support

I use an Air for ML. It's perfect if you work in the cloud. Locally it's too much hassle. Not just because of the limited power, but because of lack of support for MPS in recent kernels in modern frameworks.