Retrieval-augmented generation processor

Introduced 2.12

The retrieval_augmented_generation processor is a search results processor that you can use in conversational search for retrieval-augmented generation (RAG). The processor intercepts query results, retrieves previous messages from the conversation from the conversational memory, and sends a prompt to a large language model (LLM). After the processor receives a response from the LLM, it saves the response in conversational memory and returns both the original OpenSearch query results and the LLM response.

As of OpenSearch 2.12, the retrieval_augmented_generation processor supports only OpenAI and Amazon Bedrock models.

Request body fields

The following table lists all available request fields.

FieldData typeDescription
model_idStringThe ID of the model used in the pipeline. Required.
context_field_listArrayA list of fields contained in document sources that the pipeline uses as context for RAG. Required. For more information, see Context field list.
system_promptStringThe system prompt that is sent to the LLM to adjust its behavior, such as its response tone. Can be a persona description or a set of instructions. Optional.
user_instructionsStringHuman-generated instructions sent to the LLM to guide it in producing results.
tagStringThe processor’s identifier. Optional.
descriptionStringA description of the processor. Optional.

Context field list

The context_field_list is a list of fields contained in document sources that the pipeline uses as context for RAG. For example, suppose your OpenSearch index contains a collection of documents, each including a title and text:

  1. {
  2.   "_index": "qa_demo",
  3.   "_id": "SimKcIoBOVKVCYpk1IL-",
  4.   "_source": {
  5.     "title": "Abraham Lincoln 2",
  6.     "text": "Abraham Lincoln was born on February 12, 1809, the second child of Thomas Lincoln and Nancy Hanks Lincoln, in a log cabin on Sinking Spring Farm near Hodgenville, Kentucky.[2] He was a descendant of Samuel Lincoln, an Englishman who migrated from Hingham, Norfolk, to its namesake, Hingham, Massachusetts, in 1638. The family then migrated west, passing through New Jersey, Pennsylvania, and Virginia.[3] Lincoln was also a descendant of the Harrison family of Virginia; his paternal grandfather and namesake, Captain Abraham Lincoln and wife Bathsheba (née Herring) moved the family from Virginia to Jefferson County, Kentucky.[b] The captain was killed in an Indian raid in 1786.[5] His children, including eight-year-old Thomas, Abraham's father, witnessed the attack.[6][c] Thomas then worked at odd jobs in Kentucky and Tennessee before the family settled in Hardin County, Kentucky, in the early 1800s.[6]\n"
  7.   }
  8. }

You can specify that only the text contents should be sent to the LLM by setting "context_field_list": ["text"] in the processor.

Example

The following example demonstrates using a search pipeline with a retrieval_augmented_generation processor.

Creating a search pipeline

The following request creates a search pipeline containing a retrieval_augmented_generation processor for an OpenAI model:

  1. PUT /_search/pipeline/rag_pipeline
  2. {
  3.   "response_processors": [
  4.     {
  5.       "retrieval_augmented_generation": {
  6.         "tag": "openai_pipeline_demo",
  7.         "description": "Demo pipeline Using OpenAI Connector",
  8.         "model_id": "gnDIbI0BfUsSoeNT_jAw",
  9.         "context_field_list": ["text"],
  10.         "system_prompt": "You are a helpful assistant",
  11.         "user_instructions": "Generate a concise and informative answer in less than 100 words for the given question"
  12.       }
  13.     }
  14.   ]
  15. }

copy

Using a search pipeline

Combine an OpenSearch query with an ext object that stores generative question answering parameters for the LLM:

  1. GET /my_rag_test_data/_search?search_pipeline=rag_pipeline
  2. {
  3.   "query": {
  4.     "match": {
  5.       "text": "Abraham Lincoln"
  6.     }
  7.   },
  8.   "ext": {
  9.     "generative_qa_parameters": {
  10.       "llm_model": "gpt-3.5-turbo",
  11.       "llm_question": "Was Abraham Lincoln a good politician",
  12.       "memory_id": "iXC4bI0BfUsSoeNTjS30",
  13.       "context_size": 5,
  14.       "message_size": 5,
  15.       "timeout": 15
  16.     }
  17.   }
  18. }

copy

For more information about setting up conversational search, see Using conversational search.