BigScience released T0*, an encoder-decoder model series that outperforms GPT-3 on many zero-shot tasks while being 16x smaller. This tutorial walks through deploying the 3B parameter version (T0_3B) to Amazon SageMaker using the Hugging Face SageMaker SDK. Steps include setting up a SageMaker Notebook Instance, defining HF_MODEL_ID and HF_TASK environment variables, deploying to an ml.g4dn.2xlarge instance, and running inference via HuggingFacePredictor.
