DataPulse, an analytics company, wants to fine-tune a foundation model in Amazon Bedrock to improve responses on its product-support dialogues. You have prepared a training file and uploaded it to s3://datapulse-bedrock/train/support_finetune.jsonl. Bedrock expects one training example per line. Which of the following is the correct preparation and submission workflow to create a Bedrock fine-tuning (model customization) job?

The correct answer outlines the precise steps for fine-tuning a foundation model in Amazon Bedrock. Bedrock's CreateModelCustomizationJob expects training data in JSONL format, where each line is a JSON object with "prompt" and "completion" keys. The trainingDataS3Uri must point to the specific JSONL file in S3, and an IAM role with S3 read permissions is required. The trainingFormat must be explicitly set to "jsonl". Incorrect options: Bedrock does not use SageMaker HyperParameterTuningJob for fine-tuning foundation models. Bedrock expects the full S3 object path for trainingDataS3Uri, not just the bucket name, and the training format for this use case is JSONL, not CSV. Bedrock requires "prompt" and "completion" keys, not "instruction" and "response", and the file should not be wrapped in a top-level JSON object. An IAM role is always required for Bedrock to access S3 resources.