AWS – Amazon Bedrock Model Distillation is now generally available
Model Distillation is the process of transferring knowledge from a more capable model (teacher) to a less capable one (student) with the goal to make the faster and cost-efficient student model as performant as the teacher for a specific use-case. With general availability, we now add support for the following new models: Amazon Nova Premier (teacher) and Nova Pro (student), Claude 3.5 Sonnet v2 (teacher), Llama 3.3 70B (teacher) and Llama 3.2 1B/3B (student). Amazon Bedrock Model Distillation now enables smaller models to accurately predict function calling for Agents use cases while helping to deliver substantially faster response times and lower operational costs. Distilled models in Amazon Bedrock are up to 500% faster and 75% less expensive than original models, with less than 2% accuracy loss for use cases like RAG. In addition to RAG use cases, Model Distillation also adds support for data augmentation for Agents use cases for function calling prediction.
Amazon Bedrock Model Distillation offers a single workflow that automates the process needed to generate teacher responses, adds data synthesis to improve teacher responses, and then trains the student model. Amazon Bedrock Model Distillation may choose to apply different data synthesis methods that are best suited for your use-case to create a distilled model that approximately matches the advanced model for the specific use-case.
Learn more in our documentation, website, and blog.
Read More for the details.