AWS SageMaker

SageMaker can be used for both chat and embedding models. Chat models are supported for endpoints deployed with LMI, and embedding models are supported for endpoints deployed with HuggingFace TEI

Here is an example Sagemaker configuration setup:

YAML

config.yaml

models:
  - name: deepseek-6.7b-instruct
    provider: sagemaker
    model: lmi-model-deepseek-coder-xxxxxxx
    region: us-west-2
    roles:
      - chat
  - name: mxbai-embed
    provider: sagemaker
    model: mxbai-embed-large-v1-endpoint
    roles:
      - embed

JSON

config.json

{
  "models": [
    {
      "title": "deepseek-6.7b-instruct",
      "provider": "sagemaker",
      "model": "lmi-model-deepseek-coder-xxxxxxx",
      "region": "us-west-2"
    }
  ],
  "embeddingsProvider": {
    "provider": "sagemaker",
    "model": "mxbai-embed-large-v1-endpoint"
  }
}

The value in model should be the SageMaker endpoint name you deployed.

Authentication will be through temporary or long-term credentials in ~/.aws/credentials under a profile called "sagemaker".

[sagemaker]
aws_access_key_id = abcdefg
aws_secret_access_key = hijklmno
aws_session_token = pqrstuvwxyz # Optional: means short term creds.