Optimum Neuron Container

We provide pre-built Optimum Neuron containers for Amazon SageMaker. These containers come with all of the Hugging Face libraries and dependencies pre-installed, so you can start using them right away. We have containers for training and inference, and optimized text generation containers with TGI. The table is up to date and only includes the latest versions of each container. You can find older versions in the Deep Learning Container Release Notes

It is possible to ure the get_huggingface_llm_image_uri function from the sagemaker Python SDK to retrieve the Text Generation Inference URI for the container you want to use.

from sagemaker.image_uri_config import get_huggingface_llm_image_uri

llm_image = get_huggingface_llm_image_uri("huggingface-neuronx")

If you have the Optimum Neuron package installed, you can use the function image_uri to retrieve the image URI for the container you want to use. The result is the same as the one retrieved by the sagemaker Python SDK, but the image URI retrieved can be newer than the one reported by the sagemaker Python SDK.

from optimum.neuron.utils import ecr

# retrieve the llm image uri
llm_image = ecr.image_uri("tgi")

print(f"llm image uri: {llm_image}")

Available Optimum Neuron Containers

Please replace 763104351884 with the correct AWS account ID and region with the AWS region you are working in.