Summarizing Text with H2O Label Genie
- Generative AI
- H2O Cascade
This lesson walks through setting up a text summarization task in H2O Label Genie using a sample of CNN Daily Mail news articles. It covers naming the task, selecting the dataset and the text column to summarize, and choosing among available models such as Facebook's BartLargeCNN, Google's PegasusLarge, H2O GPT, and OpenAI's GPT-3.5 and GPT-4. The lesson finishes by reviewing the zero shot generated summaries side by side with the original articles, editing any summary that needs adjustment, and saving each one to move to the next.
Key takeaways
- How to configure a summarization task in Label Genie by choosing a dataset and a text column
- What summarization models are available in Label Genie, from BartLargeCNN to H2O GPT and GPT-4
- Why zero shot predictions still need a human review pass before a summary is saved and approved
- How editing a generated summary directly in Label Genie fits into the annotation workflow
Start owning your AI