Error loading model from cache in docker container

[skolachi]

I am unable to load the model from cache in offline mode.

1. I set environment variable HF_HOME to /.cache/huggingface/hub in the Dockerfile.
2. I copied the cached models from my local to HF_HOME in the Dockerfile. GLiNER model requires the mdeberta model so I copied that model as well to HF_HOME.
3. In the container, I load the model from cache using the following Python code-

from gliner import GLiNER
import is
model = GLiNER.from_pretrained(“urchade/gliner_multi-v2.1”, cache_dir=os.environ[“HF_HOME”], local_files_only=True)

This throws the following error-

requests.exceptions.SSLError: (MaxRetryError("HTTPSConnectionPool(host='huggingface.co', port=443): Max retries exceeded with url: /microsoft/mdeberta-v3-base/resolve/main/tokenizer_config.json (Caused by SSLError(SSLCertVerificationError(1, '[SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed: unable to get local issuer certificate (_ssl.c:1123)')))"), '(Request ID: 2deb8e47-d667-42c7-9d03-b7afdac14b88)')

It looks like the code is trying to download mdeberta model again.

I also tried to set cache_dir to other options such as “/.cache/huggingface”, “/.cache/huggingface/hub/models—urchade—gliner_multi-v2.1”, “/.cache/huggingface/hub/models—urchade—gliner_multi-v2.1/snapshots” and “/.cache/huggingface/hub/models—urchade—gliner_multi-v2.1/snapshots/”.

All of these options throw the following error-

huggingface_hub.utils._errors.LocalEntryNotFoundError: Cannot find an appropriate cached snapshot folder for the specified revision on the local disk and outgoing traffic has been disabled. To enable repo look-ups and downloads online, pass 'local_files_only=False' as input.

I am able to load the cached model using the code in step 3 on my local machine but not in the docker container.

I am using gliner version 0.2.8 and bullseye Debian.

Anyone else run into this? Any help is appreciated.