load_file errors when device is `mps:0`

[jakelorocco]

System Info

• transformers version: 4.53.3
• Platform: macOS-15.6-arm64-arm-64bit
• Python version: 3.12.0
• Huggingface_hub version: 0.36.0
• Safetensors version: 0.6.2
• Accelerate version: 1.11.0
• Accelerate config: not found
• DeepSpeed version: not installed
• PyTorch version (accelerator?): 2.7.0 (NA)
• Tensorflow version (GPU?): not installed (NA)
• Flax version (CPU?/GPU?/TPU?): not installed (NA)
• Jax version: not installed
• JaxLib version: not installed
• Using distributed or parallel set-up in script?: no

Information

• The official example scripts
• My own modified scripts

Reproduction

from safetensors.torch import load_file
load_file('.../adapter_model.safetensors', device="mps:0") # any adapter_model.safetensors file

Results in:

E       safetensors_rust.SafetensorError: device mps:0 is invalid

Expected behavior

I encounter this issue when loading adapters using Peft, but I believe this is a generic issue with safetensors. I think the expected behavior is that safetensors should load this file using mps. Torch will seems to accept torch.device("mps:0") and will accept .to("mps:0") when moving tensors to a device.