RAG¶

class rag.rag_processor.RAGProcessor(llm: BaseLLM, max_token: int, temperature: float, top_p: float, retriever, top_k: int)¶

Class implementing the RAG process

Parameters:

  • llm (BaseLLM)

  • max_token (int)

  • temperature (float)

  • top_p (float)

  • retriever

  • top_k (int)

init_retriever_client(retrieval_method: str = 'top_k')¶

Initialize and return a retriever client based on retrieval_method.

Returns:

An instance of the appropriate retriever client based on retrieval_method.

Return type:

object

create_rag_message(context_docs: List[Any], query: str) → List[Dict]¶

Format the RAG message to send to the OpenAI API.

Parameters:

  • context_docs (str) – Context matching the query according to the retrieval process

  • query (str) – User input question

Returns:

Contains the message in the correct format to send to the OpenAI API

Return type:

list of dict

retrieve(db: Session, request: RAGRequest, language: str = None, tag: str = None, k: int = 0)¶

Retrieve context documents related to the user input question.

Parameters:

  • db (Session) – Database session

  • request (RAGRequest) – User input question

  • language (str) – Question and context documents language

  • tag (str) – Tag to filter the context documents

  • k (int, default 0) – Number of context documents to return

process(db: Session, request: RAGRequest, language: str = None, tag: str = None)¶

Process a RAGRequest to retrieve relevant documents and generate a response.

This method retrieves relevant documents from the database, constructs a context from the documents, and then uses an LLM client to generate a response based on the request query and the context.

Parameters:

  • db (Session) – The database session to use for retrieving documents.

  • request (RAGRequest) – The request to process.

  • language (str, optional) – The language of the documents to retrieve. If not specified, documents in all languages are considered.

  • tag (str, optional) – The tag to filter the documents to retrieve. If not specified, all documents are considered.

Returns:

The response generated by the LLM client.

Return type:

str

async embed(text_input: EmbeddingRequest)¶

Get the embedding of an embedding request.

Parameters:

text_input (EmbeddingRequest)

Returns:

The requested text embedding

Return type:

dict