A company wants to customize Amazon Bedrock foundation models (FMs) to improve an application's performance. The company must prepare a training dataset for text-to-text model fine-tuning. Which…

For fine-tuning text-to-text foundation models on Amazon Bedrock, the training dataset typically needs to be in a JSON Lines format, where each line is a JSON object containing distinct fields for prompts and corresponding completions (labeled data). This structured format allows the model to learn specific input-output relationships.

Unlabeled data is used for pre-training or unsupervised learning, not for fine-tuning where specific input-output pairs are needed to adapt the model to a particular task.

While CSV files can store tabular data, for text-to-text fine-tuning, a JSON-based format (like JSON Lines) with explicit prompt/completion fields is generally required by platforms like Amazon Bedrock to clearly delineate the input and target outputs.

A simple text file with unlabeled data is too unstructured and lacks the necessary prompt-completion pairing required for supervised fine-tuning of text-to-text models.