RAGWorkflow
DashAI.back.workflows.RAGWorkflow
Retrieval-Augmented Generation over the session's documents.
Runs the text to text task: the session model is a text to text LLM and each turn retrieves chunks of the session's documents before the LLM answers. The chunkers, retrievers, generation prompts and query transformations a RAG session can use are the ones related here, so a workflow over other content (images, say) relates its own.
Methods
process_output(self, output: 'RAGGenerationOutput', **kwargs: Any) -> List[Tuple[str, str]]
RAGWorkflowTurn a pipeline output into the five typed output rows.
Parameters
- output : RAGGenerationOutput
- The typed output of
RAGPipeline.generate(). - **kwargs : Any
- Unused; accepted so callers can pass the generic job arguments.
Returns
- List[Tuple[str, str]]
(data, type)rows ordered Queries, Dict, Trace, Timings, str.
Compatible with
TextToTextGenerationTaskCharacterChunkModelRecursiveCharacterChunkModelTokenChunkModelBM25RetrieverTFIDFRetrieverDenseEmbeddingRetrieverSentenceTransformerCrossEncoderRetrieverMMRRerankerRetrieverSequentialRetrieverParallelRetrieverAgenticSingleRetrieverAgenticEnsembleRetrieverAgenticRerankerRetrieverDefaultRAGGenerationPromptCustomRAGGenerationPromptDefaultQARAGGenerationPromptLLMQueryRewriterLLMQueryExpanderLLMQueryDecomposer