Operationalization of AI/ML models and use cases for ChatGPT Q&A
summary
Operationalizing LLMs focuses on concrete choices: use Prompt Flow and retrieval-augmented generation (RAG) for fast, cheap prototypes; fine-tune when prompts and RAG can’t reach required accuracy, often with few hundred high-quality examples for GPT‑3.5-class models; store fine-tune checkpoints and training data in customer tenant storage; monitor with business-specific nightly metrics to detect drift; chain open-source models like Whisper or Hugging Face for multimodal steps and use LoRA-style deltas to avoid cloning full base weights.
AI-written summary · about
chapters (13)
- 0:00Session introduction
- 2:54Productionizing LLMs
- 5:58Data lake vs LLMs
- 8:44Data quality and monitoring
- 12:09Fine-tuning logistics
- 15:26Prompt engineering vs tuning
- 18:31Multimodal applications
- 22:12Rag and fine-tuning strategies
- 26:06Open source and hybrid models
- 29:29Governance and security
- 34:25Model updates and refreshers
- 38:22Data ownership and monitoring
- 44:23Session wrap-up
related talks
27mWHAT DOES MLOps LOOK LIKE FOR LARGE LANGUAGE MODELS (LLMs) IN PRODUCTION?
51mChatGPT as a developer's personal servant
30mAI Application Architecture: Building for Scale and Efficiency
46mConversations with ChatGPT: illusion or reality?
44mExploring ChatGPT for Improved Observability
21mChatGPT, LLMs, and LangChains - A Beginner's Guide
48mDeploy large language models responsibly with Azure AI
18mGenerative AI in Financial Industry