v0.2.14
📦 ragas
✨ 6 features🐛 8 fixes🔧 4 symbols
Summary
This release introduces new features like HTTP request-response logging and multi-turn conversation evaluation, alongside numerous bug fixes across various metrics and synthesizers. It also includes documentation updates and new integrations.
✨ New Features
- Added HTTP request-response logging controlled by an environment flag.
- Enabled evaluation for Multi-Turn Conversations.
- Introduced NVIDIA end-to-end accuracy, relevance, and groundedness metrics.
- Enabled Runtime Checks for ModeMetric Protocol.
- Added r2r integration.
- Added Haystack LLM and embedding wrapper.
🐛 Bug Fixes
- Resolved issue where metrics with different parameters yielded different scores.
- Fixed simple criteria metric calculation.
- Fixed SingleHopQuerySynthesizer::prepare_combinations().
- Improved Groundedness metric by implementing 5 retries with early break.
- Ensured only the node ID is saved in relationships when saving knowledge graphs (kg).
- Fixed return value to be valid JSON output schema.
- Updated default.py to include missing `cosine_similarity` transform for documents with Token Count 101–500.
- Fixed ndarray dtype issues and improved error wording.
🔧 Affected Symbols
StatementGeneratorPromptSingleHopQuerySynthesizer::prepare_combinationsModeMetric Protocoldefault.py