Syllaro Academy

AWS Certified AI Practitioner

Preparing Data for Fine-Tuning

Fine-tuning succeeds or fails on its data. This lesson covers the JSONL formats Bedrock expects, how much data you need, and what makes a dataset clean, representative, and safe to train on.

Intermediate 17 minutes 4 Learning Objectives
  1. Write the JSONL formats Bedrock expects for supervised fine-tuning and for continued pre-training
  2. Explain the role of a validation dataset and the training-and-validation split
  3. Identify what makes training data high quality: representative, balanced, clean, and consistent
  4. Recognize the governance and bias risks in a fine-tuning dataset