Add HF authentication instructions to lorax-launcher docs about lorax

[tgaddair]

Hey @binarycrayon, for adapters you can pass in the api_token param in the request for now. Although setting a token in the environment should apply that to all requests by default.

[JamsheedMistri]

Ah yes, I only tested with a gated model, not a gated adapter.

[binarycrayon]

Did you verify this approach work?
I am getting 401 Client Error from lorax server when trying to retrieve gated lora, with
-e HUGGING_FACE_HUB_TOKEN=$HF_TOKEN

[JamsheedMistri]

This code works for me:

envs:
  MODEL_ID: my_model
  HUGGING_FACE_HUB_TOKEN: my_token

run: |
  docker run --gpus all \
    --shm-size 1g \
    -p 8080:80 \
    -e HUGGING_FACE_HUB_TOKEN=$HUGGING_FACE_HUB_TOKEN \
    ghcr.io/predibase/lorax:latest \
    --model-id $MODEL_ID

If I remove the -e flag it no longer works.

[binarycrayon]

thanks for confirming, I'll double check my token and my setup

[binarycrayon]

it is still failing for me. Downloading a gated model work, but loading a gated lora does not.

Looks like it's failing at

lorax/server/lorax_server/adapters/utils.py

Line 20 in 157c97d

HfApi(token=api_token).model_info(adapter_id, revision=None)

I'm trying to figure out where they get the auth token from