As the huggingface-inference-toolkit version was pinned to a more recent one (0.4.2) instead of 0.4.1, that's due to the version pinning on huggingface-inference-toolkit that happened before releasing it publicly.
Additionally, this PR also adds the exec before the uvicorn command in the entrypoint.sh files so as to fix the signal handling for gracefully shutting it down.
Finally, this PR also installs the correct version of huggingface_hub before installing the huggingface-inference-toolkit to prevent the latest from being installed, which deprecates the cached_download method used by other dependencies.
Description
This PR fixes the version of
huggingface-inference-toolkitpinned within both DLCs:As the
huggingface-inference-toolkitversion was pinned to a more recent one (0.4.2) instead of 0.4.1, that's due to the version pinning onhuggingface-inference-toolkitthat happened before releasing it publicly.Additionally, this PR also adds the
execbefore theuvicorncommand in theentrypoint.shfiles so as to fix the signal handling for gracefully shutting it down.Finally, this PR also installs the correct version of
huggingface_hubbefore installing thehuggingface-inference-toolkitto prevent the latest from being installed, which deprecates thecached_downloadmethod used by other dependencies.