how to load jina-clip-implementation by local path ?
#16
by lzw1992 - opened
I have downloaded the model weights for jinaai/jina-clip-v2 on a machine with internet access. Now, I need to run jinaai/jina-clip-v2 on a server that has no internet connection. When loading the model, it tries to download jina-clip-implementation. How can I specify a local path for jina-clip-implementation to avoid downloading it?
Hey @lzw1992 ! You can try the following:
- Clear your HF cache
rm -rf ~/.cache/huggingface/ - Download the model from huggingface hub
python -c "from transformers import AutoModel;AutoModel.from_pretrained('jinaai/jina-clip-v2', trust_remote_code=True)" - Now your cache folder will contain everything required to load the model, code + weights and nothing else
- Copy your cache folder to the server
scp -r ~/.cache/huggingface/ user@host:/home/user/.cache/ - Try loading on the server
python -c "from transformers import AutoModel;AutoModel.from_pretrained('jinaai/jina-clip-v2', trust_remote_code=True)"
Thanks, I got it !
When I download the model file and run the model locally, do I need other pre-trained models?
@zhanbaohang no you don't need, this repo contains the pre-trained model weights of jina-clip, this includes both text tower and image tower.
gmastrapas changed discussion status to closed