Closed fakerybakery closed 1 year ago
I haven't used MPS before, will investigate it.
After some attempts, I found that setting device="mps"
alone is not enough, and we also need an additional model.to("mps")
to run.
What's even stranger is that the inference speed (of distillgpt2
) actually decreases significantly when using MPS compared to using CPU (Macbook Pro M1).
OK, thank you. I will try that out.
Hello, Thank you for making this great repository! Is it possible to run this on M1/M2 chips using MPS? I've tried setting
self.device
tomps
, however I get this:Is there any way to run this using MPS optimization? Thank you!