Hello,
there is no Split_Einsum SD3 or Flux.1 models for ANE. I asked several developers, what is the reason for this, without an answer!
I have a MacBook Pro M2 16gb RAM and an iPad M4 16gb RAM with a stunning powerful Neural Engine that i cannot use and i don't have 72 GPU cores of a M2 Ultra to run the ORIGINAL versions fast, shouldn't it be the priority to improve the ANE support because those are ML models and we have very powerful ANE in our little machines, which is made exactly for this - to run those models!
can someone of the developers from Apple or ArgMax explain and give a consistent answer, why SDXL 1024x1024 or newer models like Flux.1 cannot be compiled for CoreML ANE?
Hello,
there is no Split_Einsum SD3 or Flux.1 models for ANE. I asked several developers, what is the reason for this, without an answer!
I have a MacBook Pro M2 16gb RAM and an iPad M4 16gb RAM with a stunning powerful Neural Engine that i cannot use and i don't have 72 GPU cores of a M2 Ultra to run the ORIGINAL versions fast, shouldn't it be the priority to improve the ANE support because those are ML models and we have very powerful ANE in our little machines, which is made exactly for this - to run those models!
can someone of the developers from Apple or ArgMax explain and give a consistent answer, why SDXL 1024x1024 or newer models like Flux.1 cannot be compiled for CoreML ANE?