News for hackers. Stuff that matters.
Re:New Mac Mini with M4 by burnerthrow008 on Tuesday October 29, 2024 @11:59AM (#41985546)
Lack of Cuda is not a problem if for most ML frameworks. For example, in PyTorch you just tell it to use the “mps” (metal performance shaders) device instead of the “cuda” device.
Re:New Mac Mini with M4 by xattt on Tuesday October 29, 2024 @12:11PM (#41985747)
Cuda Apple license it from nVidia?
Re:New Mac Mini with M4 by rob74 on Tuesday October 29, 2024 @12:18PM (#41985887)
Cude, er, cute, but... no.
Re:New Mac Mini with M4 by ojhughes on Tuesday October 29, 2024 @12:45PM (#41986344)
I tried training some models using tensorflow-metal a year ago and I was quite disappointed. Using a relu activation function led to very poor accuracy [0] and training time was an order of magnitude slower than just using the free tier of Google Colab

[0] https://github.com/keras-team/tf-keras/issues/140

Re:New Mac Mini with M4 by throwaway314155 on Tuesday October 29, 2024 @02:48PM (#41988016)
That simply isn't true in practice. Maybe for inference, but even then you're running up against common CUDA kernels such as FlashAttention which will be far from plug and play with PyTorch.