Apple mps backend support

[callankuo]

I use Mac M3 chip and test the Predict module got RuntimeError: Placeholder storage has not been allocated on MPS device!

[callankuo]

Is this means that we can't use Mac book pro for DSPY development? ( I use meta-llama/Llama-2-13b-chat-hf model already cache on my local machine.)

[setren]

Got the same issue, any updates 🤔

[callankuo]

Use Ollama framework will works.

import dspy
import torch

Apple device MPS support

device = torch.device("mps")
llama = dspy.OllamaLocal("llama3:8b", max_tokens=5000)
colbertv2_wiki17_abstracts = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')
dspy.settings.configure(lm=llama, rm=colbertv2_wiki17_abstracts, device_map = device, trace=[], temperature=0.7)

[setren]

It works, cool 🙌