### sagemaker-gemma

Serving `google/gemma-4-E2B-it` with vLLM on a **SageMaker real-time endpoint**, managed
entirely through the aws CLI.

| Setting | Value |
| --- | --- |
| Region | `us-east-2` |
| Endpoint | `gemma-4-e2b` |
| Instance types (priority order) | `ml.g6.xlarge`, `ml.g6.2xlarge`, `ml.g6.4xlarge` |
| Max model length | `8192` |
| Image | `newest SageMaker vLLM image (find_vllm_image)` |
| Execution role | `sagemaker-gemma-execution-role` |

**Order of work:** check_quotas → deploy_endpoint → get_endpoint_status until
`InService` (about 10 minutes once an instance is placed) → verify_model_health
→ query_model → delete_endpoint.