Amazon Bedrock Model Distillation is now generally available

Amazon Bedrock Model Distillation is now generally available, offering a method to transfer knowledge from advanced (teacher) to smaller (student) models with improved performance and efficiency.

• Supports new models: Nova Premier/Pro, Claude 3.5 Sonnet v2, Llama 3.3 70B/3.2 1B/3B
• Enables smaller models to accurately predict function calling for Agents use cases
• Distilled models are up to 500% faster and 75% less expensive
• Maintains less than 2% accuracy loss for use cases like RAG
• Provides automated workflow for generating, synthesizing, and training models
• Supports data augmentation for improved function calling prediction

The service aims to make AI models more cost-effective and performant by intelligently transferring knowledge from advanced to smaller models.