H2O Generative AI Ecosystem Overview
- Generative AI
Overview
This video is the closing overview in H2O's generative AI ecosystem series, zooming out from individual components to show how the pieces fit together end to end. It walks through the full path from raw unstructured documents to a working GenAI application: preparing data in LLM Data Studio, fine-tuning a model in LLM Studio, adding retrieval-augmented generation through Enterprise h2oGPT, and evaluating models with Eval Studio. Live product demos show each stage in action, including uploading a PDF and running a rag-based chat.
Key takeaways
- How H2O frames the foundation of a generative AI ecosystem: unstructured documents, data preparation, fine-tuning or prompt tuning, and app-level API access
- How LLM Data Studio curates, cleans, and augments data, including removing profanity and sensitive information and filtering low quality auto-generated question-answer pairs
- How to configure a fine-tuning experiment in LLM Studio, including selecting a model backbone, tokenizer settings, quantization precision, and LoRA hyperparameters
- How Enterprise h2oGPT stores documents in a collection and performs retrieval-augmented generation, including uploading a PDF, configuring rag prompts, and generating an API key
- How Eval Studio and the public eval.gpt leaderboard track open source model performance with an ELO rating system
- How the four tools, Data Studio, LLM Studio, Enterprise h2oGPT, and Eval Studio, connect into one workflow from raw documents to a deployed GenAI app
Start owning your AI