Issue loading private Transformer model in version 3.3.0

[J-Curwell]

Issue:

We have a fine-tuned version of ember-v1 (https://huggingface.co/llmrails/ember-v1) hosted as a private model in our Hugging Face org. We could successfully load this model in the previous release of sentence-transformers (3.2.1) but as of 3.3.0 this same code no longer works.

Steps to reproduce:

• Install sentence-transformers 3.3.0
• Ensure you do not have the HF_TOKEN environment variable set. Our architecture reads this directly from a secret manager rather than setting as an environment variable
• Run:

  
  from sentence_transformers.models.Transformer import Transformer

args = { "token": , "trust_remote_code": False, "revision": None, "local_files_only": False }

transformer = Transformer( model_name_or_path=, cache_dir=None, backend="torch", max_seq_length=512, do_lower_case=True, model_args=args, tokenizer_args=args, config_args=args )

This same code worked before. I believe I have isolated the issue to this recent PR #2980 . Specifically the call to the `find_adapter_config_file` on new line 104, which does not pass in the auth token.

Is the suggested approach to always set the `HF_TOKEN` environment variable?

[HenningDinero]

Can confirm - have the exact same issue. I have tried parsing the tokens directly and that doesn't work.

If I downgrade to 3.2.1 then it works like a charm

[tomaarsen]

Feel free to stick with 3.2.1 for now, unless you'd like to try out some of the new v3.3.0 features like Training with Prompts, NanoBEIR Evaluation, PEFT, Int8 quantization with OpenVINO, etc. In that case, you can download the bleeding edge version until I bring out v3.3.1:

pip install git+https://github.com/UKPLab/sentence-transformers.git

Thanks for reporting this!

• Tom Aarsen