Currently, the Translation pipeline loads a list of supported translation models in the __init__ method. This is causing HTTP 429 errors from the Hugging Face Hub with the build script.

The pipeline should load the list of supported models once on demand.