Version 0.2.4 - 2025-09-18
Added
Added
DocumentSynthesizerfor document text extraction and chunking, enabling the creation of context from documents.Added
ContextGeneratorservice with semantic chunking for improved context selection.Added support for Ollama LLM provider.
Added document source tracking to
DocumentSynthesizer.Added
strategyparameter toDocumentSynthesizerfor sequential vs random context selection.Added comprehensive user feedback for test generation plan.
Changed
Refactored benchmarking framework to integrate SDK modules and improve model handling.
Refactored
PromptSynthesizerto use context instead of documents.Refactored
RhesisPromptMetricfor improved performance and maintainability.Updated
DocumentSynthesizerto useDocumentdataclass instead of dictionaries.Updated LLM providers to inherit from LiteLLM for GeminiLLM and OpenAILLM.
Improved model factory for easier model selection and configuration.
Metrics now accept a model directly, allowing for more flexible model integration.
Replaced
ContextSynthesizerwithContextGeneratorservice.Enforced hard size limits with semantic-first splitting in
ContextGenerator.
Fixed
Fixed batch size None comparison error.
Fixed testset generation issues.
Fixed minor issues in LLM provider implementations.
Fixed template path for metrics.
Fixed Python package version conflicts.
Removed
Removed
pyarrowdependency to reduce environment and Docker image sizes.Removed the template caching mechanism.
Removed binary and categorical functionality from prompt metrics.
Removed the absolute_max_context_tokens limit.
Removed document support from
PromptSynthesizer.Removed the need for API keys in some configurations.