Model choice in Amazon Bedrock
summary
Amazon Bedrock exposes models from AI21 Labs, Anthropic, Cohere, Meta, Mistral, Stability and Amazon via a single API and adds managed RAG (ingest PDFs → embeddings → vector DB → retrieval with citations), a fully managed agents service for task decomposition and API actions, and custom model import plus fine‑tuning (LoRA recommended for low data). Context windows range ~4k–200k tokens. Built‑in model evaluation runs automated and human‑in‑the‑loop tests — example: Titan text Express scores 0.21 accuracy and ~0.968 on a WikiText2 robustness split.
AI-written summary · about
chapters (9)
- 0:00Introduction to model choice
- 4:57Intrinsic model properties
- 15:45Initial model selection
- 17:33Prompt engineering techniques
- 19:57Retrieval augmented generation
- 23:29Building agents for workflows
- 27:20Fine-tuning models
- 30:32Model evaluation strategies
- 37:56Security and scalability
related talks
39mEvaluate and choose models in Amazon Bedrock (AIM204)
29mCustomizing genAI applications for your business
55mSecurely build generative AI apps & control data with Amazon Bedrock
43mBeyond Chatbots: How Amazon Bedrock Agents Drive Enterprise AI
24mBuilding GenAI applications with Amazon Bedrock
31mLamRAG: From Data to Constructive Insights using Amazon Bedrock
22mObservability in AI-Powered Applications with Amazon Bedrock
29mAnthropic's Claude in Amazon Bedrock