Fix pinned huggingface-inference-toolkit to include path patch

[alvarobartt]

Description

As a follow up of #124, this PR updates the pinned version so that it points to huggingface/huggingface-inference-toolkit#76 instead, as it contains a fix for the Google Cloud Storage (GCS) path downloads.

To reproduce locally one can run the following:

Warning

The command below assumes that huggingface-pytorch-inference-cpu.2.2.2.transformers.4.41.0.py311 is the local Docker image name built as:

docker build --platform=linux/amd64 -t huggingface-pytorch-inference-cpu.2.2.2.transformers.4.41.0.py311 -f containers/pytorch/inference/cpu/2.2.2/transformers/4.41.1/py311/Dockerfile .

And that the AIP_STORAGE_URI is pointing to a valid bucket on GCS, with a Transformers compatible model.

docker run --platform linux/amd64 -v ~/.config/gcloud:/root/.config/gcloud -e GOOGLE_APPLICATION_CREDENTIALS="/root/.config/gcloud/application_default_credentials.json" -e GOOGLE_CLOUD_PROJECT="yout-project-id" -e HF_TASK="your-task" -e AIP_STORAGE_URI="your-bucket-path" huggingface-pytorch-inference-cpu.2.2.2.transformers.4.41.0.py311