Make HUGGINGFACE_OFFLINE configurable

[slyt]

Currently HUGGINGFACE_OFFLINE=1 is hardcoded in the helm template.

I'd like for it to be configurable in values.yaml so that I don't have to pull models from s3 and can download them directly from huggingface hub via the functionality in huggingface text-generation-inference image.

[slyt]

I was able to overwrite it with:

extraEnvVars:
  - name: HUGGINGFACE_OFFLINE
    value: "0"

[voatsap]

Will take a look on this @romanprog please check

[voatsap]

don't have to pull models from s3 and can download them directly from huggingface hub via the functionality in huggingface text-generation-inference image.

There is another block that refered on if model is downloaded through s3 or directly from HF
init:
s3:
enabled: false
bucketURL: s3://k8s-model-zephyr/llm/deployment/HuggingFaceH4/zephyr-7b-beta

If s3 enabled: false it would download it directly from HF, if not, it would use bucketURL