0.20.0 Release Notes

This version includes new history field, enhancements to llm properties and improved explainability highlighting, along with more features, demos, stability and performance improvements, that are part of our 0.20.0 release.

Deepchecks LLM Evaluation 0.20.0 Release

• 💬 New “History” field
• 🏦 LLM properties bank enhancements
• 🟣 Multiple line highlighting for property explainability
• 🍿 Use case demos: Classification and Guardrails
• 📩 Data logging: partial interaction logging, steps download and upload

What’s New and Improved?

• New special field: History

  • For supplying previous historical context, such as chat history. Relevant properties will use the “History” field as additional context for checking property values.

    

• LLM properties bank enhancements

  • Added new prompts and improved prompt performance. Includes unifying the “Completeness” prompt template into one (Non-Q&A use cases have the “coverage” built-in property for uncovering issues such as a non-complete summary).

    

• Multiple line highlighting for explainability

  • Now properties such as “Grounded in Context”, “PII”, can display more than one area attributing to the highest/lowest scores, allowing efficient RCA

    

• New demos

  • Classification Demo for working with deepchecks on a classification use casue
  • Guardrails for configuring guardrails in production

• Data logging

  • An interaction can now be gradually logged, in separate parts, useful for example for production flows: Stream Upload Documentation.
  • Interaction steps can now be downloaded and uploaded via csv and SDK