Hi all - I’ve been trying out depth estimation using vinvino02/glpn-nyu · Hugging Face on my M1 MacBook Pro. Using the CPU it works perfectly, if rather slow. If I use device=“mps” it seems to work - much faster with no errors - but the output is just noise. Am I missing something or is it broken/not implemented for MPS?
Related topics
Topic | Replies | Views | Activity | |
---|---|---|---|---|
Running transformer models on mps instead of cpu on mac | 1 | 1338 | January 18, 2025 | |
Running PyTorch + Huggingface on Apple Silicon (M1) | 1 | 1663 | August 24, 2022 | |
MPS is running slower than CPU on Mac M1 Pro | 6 | 6816 | November 1, 2022 | |
Inference on constrained devices | 0 | 293 | November 21, 2020 | |
Suggest a model to fine-tune for multi-class classification on a 32GB M1 Mac? | 1 | 427 | March 4, 2024 |