SID-1 on Amazon SageMaker — Examples

Sample code for deploying and invoking SID-1 from its AWS Marketplace model package on Amazon SageMaker.

SID-1 is a retrieval-agent model serving an OpenAI-compatible chat-completions API with tool calling and SSE streaming, running on a single NVIDIA L40S (ml.g6e.2xlarge).

Contents

• sid1-sagemaker-deploy.ipynb — end-to-end sample notebook: subscribe → deploy the endpoint → invoke (standard, streaming, and via the OpenAI-compatible route) → clean up.

Requirements

• An AWS account subscribed to the SID-1 AWS Marketplace listing.
• Service quota ml.g6e.2xlarge for endpoint usage ≥ 1 in your region.
• A SageMaker execution role.

Two settings that matter

1. InferenceAmiVersion must be al2-ami-sagemaker-inference-gpu-3-1. This selects the SageMaker host AMI with NVIDIA driver 550 — the validated pairing for this container. A different AMI can fail CUDA initialization (endpoint stuck in Creating, then a health-check failure).
2. Raise both startup timeouts to 3600 s — the model download is ~16 GB.

The notebook sets both correctly.

Documentation & support

• Prompting SID-1 (including the required tool-calling protocol): https://platform.sid.ai/docs
• Support: support@sid.ai