[watsonx docs]: after tuning or importing a model in watsonx.ai, create a deployment to get an HTTPS scoring endpoint. Pick the hardware spec that fits the model size: too small and requests time out, too large and you pay for idle capacity. Send prompts to the deployment endpoint with your IAM token. If scoring returns an error, check the deployment status first: failed deployments usually come from a bad software specification or an incompatible model format, not from the prompt.

Context: A tuned model sitting in the project is not usable until deployed.