cache offloading check is incorrect

[mobicham]

This check creates some issues with torch.compile. The type hint is bool, but in some cases, that offloading value is actually a cuda device.

Who can help?

@ArthurZucker

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
• My own task or dataset (give details below)

Reproduction

For example, with Qwen3-omni, if you do this, offloading is a cuda device, which triggers that offloading check, which crashes torch.compile:

from transformers import StaticCache
past_key_values = StaticCache(model.thinker.config, max_model_len, device, compute_dtype)
print(past_key_values.offloading)
# `cuda:0`- should be False!

Expected behavior

Static cache should not crash with torch.compile.

[Rocketknight1]

cc @Cyrilvallez who wrote that block, also see the PR at #41169!

[Cyrilvallez]

You are passing a device directly as offloading argument, that's why! See the docstring of the class in your IDE, or the signature here!

[mobicham]

My bad! I copy pasted from some old code when the offloading option was not available! Thanks!