Deploy large language models responsibly with Azure AI
summary
Shows concrete tooling and outcomes: using Azure AI Studio and the Azure Machine Learning model catalog to compare Llama 2 GPT-4 and Hugging Face models, choosing on latency and token efficiency, applying retrieval augmented generation (RAG) versus fine‑tuning (requires ~1k+ vetted Q/A), orchestrating multi‑step prompt flow workflows, deploying to managed online or private endpoints, monitoring groundedness token consumption drift and hallucinations via the responsible AI dashboard, and having rollback incident‑response and content‑safety filters in production.
AI-written summary · about
chapters (17)
- 0:00Responsible LLM deployment intro
- 3:21Microsoft’s RAI principles
- 6:14Azure AI Studio features
- 9:26Model catalog & selection
- 11:58Benchmarking & comparison
- 14:52Optimizing with RAG & fine-tuning
- 18:03Prompt engineering basics
- 23:49Fine-tuning & data prep
- 26:19Advanced Prom Flow workflows
- 30:18Secure deployment & monitoring
- 33:27Development lifecycle for LLMs
- 35:38LLM performance metrics
- 38:51RAI principles recap
- 40:17Risk & compliance planning
- 43:29Azure AI safety tools
- 46:17Training and documentation resources
- 47:22Summary
related talks
48mUsing Azure AI Foundry to manage all your Large Language Models
48mBuilding and using AI models responsibly
42mBuilding and using AI models responsibly Q&A
51mThe Next Frontier: Azure OpenAI Models and the Arrival of GPT-4
47mAccelerate your AI journey with Azure AI model catalog
38mSafe Evaluation and Rollout of AI Models
59mLearn Live: Introduction to Azure OpenAI Service
30mAI Application Architecture: Building for Scale and Efficiency