Offline mode doesn't work with models that require `trust_remote_code=True`

System Info

Google Colab:

transformers version: 4.46.3
Platform: Linux-6.1.85+-x86_64-with-glibc2.35
Python version: 3.10.12
Huggingface_hub version: 0.26.2
Safetensors version: 0.4.5
Accelerate version: 1.1.1
Accelerate config: not found
PyTorch version (GPU?): 2.5.1+cu121 (False)
Tensorflow version (GPU?): 2.17.1 (False)
Flax version (CPU?/GPU?/TPU?): 0.8.5 (cpu)
Jax version: 0.4.33
JaxLib version: 0.4.33
Using distributed or parallel set-up in script?: no

Who can help?

@Rocketknight1

Information

[ ] The official example scripts
[X] My own modified scripts

Tasks

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

Reproduction

Issue

Models that require trust_remote_code=True can't be fully saved & loaded with save_pretrained() + from_pretrained(). In offline mode on a new machine during calling from_pretrained() it doesn't locate all required files in the saved local dir and tries to reach out to hf hub for the remote code part.

How to reproduce

Colab | Kaggle Tested with popular jinaai/jina-embeddings-v3 Includes step by step reproduction + results

Additional context

Stumbled on this in Kaggle Notebook env for competition. Some Kaggle competitions require submitting code in Kaggle Notebooks, which are run later on private data and don't allow internet access. Practically, this means you must prepare all models in advance, upload them as dependencies to the submission notebook. So having transformers trying to reach out to hf-hub (when the model is already pre-downloaded) is not an option and disqualifies a group of models from usage.
Originally raised in https://github.com/UKPLab/sentence-transformers/issues/2613. Received guidance by @tomaarsen in https://github.com/UKPLab/sentence-transformers/issues/2613#issuecomment-2076964416 to overcome the issue with sentence-transformers (code snippet included)
Raising as a bug report, LMK if better to reraise as FR. Would be happy to try to contribute if confirmed

Expected behavior

Model is fully saved in local dir with save_pretrained() and can be fully loaded from a local path with from_pretrained() in offline mode

huggingface / transformers

Offline mode doesn't work with models that require `trust_remote_code=True` #34855