3 ms·
> The M1 is (supposedly) specialized for ML calculations It is, but they’re not using any of that. Only the M1 GPU at this time.
by simondotau 5y ago
> The M1 is (supposedly) specialized for ML calculations
It is, but they’re not using any of that. Only the M1 GPU at this time.
- magic_at_nodai 5y agoYeah the ANE and AMX on cpu are wrapped behind Accelerate Framework and CoreML. So you will have to use CoreML (which wasn't able to compile the latest TF BERT). ANE is also inference only. So if you want training you will have to use the GPU with Apple's Tensorflow-Metal. //part of nod.ai / SHARK team.