H2O.ai Language Services: GPTe, Fine-Tuning, and Eval
- Generative AI
- Enterprise h2oGPTe
- H2O Cascade
- H2O Eval Studio
This lesson walks through the landscape of H2O's language model tools, starting with the open source H2O GPT playground and moving to the Enterprise h2oGPTe API for chatting with document collections. It shows the Python client used to log in, query collections, and tune RAG and LLM settings, then covers verticals such as banking, digital advertising, and customer service where an AI bank assistant demo blocks a lost card end to end. The lesson also covers fine tuning a proprietary model with H2O's no code studio, pushing it to Hugging Face, deploying it, and scoring it with H2O's evaluation tooling.
Key takeaways
- How the Enterprise h2oGPTe Python client logs in, chats with a document collection, and tweaks RAG and LLM settings
- What verticals use large language models today, including loan default analysis, ad compliance checks, and SEC filing chat
- How a no code fine tuning workflow curates data, trains a model, and pushes it to Hugging Face for deployment
- Why H2O's evaluation tooling builds a test set and leaderboard to score a fine tuned model on correctness
- How H2O Label Genie and H2O Hydrogen Torch use zero shot classification and no code training for annotation and vision tasks
Start owning your AI