Generation

What is generation?

Generation is a core subprocess in our SDK that focuses on producing the final output of an RAG (Retrieval-Augmented Generation) pipeline. This output typically consists of two parts:

1. Response – the synthesized answer, typically produced by the language model.

2. References – the relevant sources used to generate the response.

To support this workflow, the GL SDK provides the following components:

1. Compressor - For packaging many passages into a single prompt context to reduce token count.

2. Context Enricher - For adding useful context (e.g. metadata) into retrieved chunks before they’re passed to the language model.

3. Deep Researcher - For performing a deep research operation within an RAG pipeline with ease.

4. Reference Formatter - For filtering and formatting the references in a clear, standardized format.

5. Relevance Filter - For filtering context chunks based on their relevance with the user query.

6. Repacker - For rearranging a list of content chunks into an order that’s more effective for downstream model consumption.

7. Response Synthesizer - For synthesizing the response based on the provided inputs and contexts.