The Generator reads a set of documents and generates an answer to a question, word by word. While extractive QA highlights the span of text that answers a query, generative QA can return a novel text answer that it has composed.
The best current approaches, such as Retriever-Augmented Generation and LFQA, can draw upon both the knowledge it gained during language model pretraining (parametric memory) as well as passages provided to it with a Retriever (non-parametric memory). With the advent of transformer-based retrieval methods such as Dense Passage Retrieval, Retriever and Generator can be trained concurrently from the one loss signal.
|Position in a Pipeline||Generator is placed after the Retriever and can be used as a substitute to the Reader.|
- More appropriately phrased answers.
- Able to synthesize information from different texts.
- Can draw on latent knowledge stored in language model.
- Not easy to track what piece of information the generator is basing its response off of.
To initialize a Generator:
from haystack.nodes import RAGeneratorgenerator = RAGenerator(model_name_or_path="facebook/rag-sequence-nq",retriever=dpr_retriever,top_k=1,min_length=2)
To run a Generator in a pipeline:
from haystack.pipelines import GenerativeQAPipelinepipeline = GenerativeQAPipeline(generator=generator, retriever=dpr_retriever)result = pipelines.run(query='What are the best party games for adults?', top_k_retriever=20)
To run a stand-alone Generator:
result = generator.predict(query='What are the best party games for adults?',documents=[doc1, doc2, doc3...],top_k=top_k)