Generator

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) and the 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.

Pros

• More appropriately phrased answers.
• Able to synthesize information from different texts.
• Can draw on latent knowledge stored in language model.

Cons

• Not easy to track what piece of information the generator is basing its response off of.

👍

Tutorial

Build your own generative QA system with RAG and LFQA.

Generator Classes

• RAGenerator: Retrieval-Augmented Generator based on Hugging Face's transformers model. Its main advantages are a manageable model size and the fact that the answer generation depends on retrieved documents. This means that the model can easily adjust to domain documents even after the training is finished.
• Seq2SeqGenerator: A generic sequence-to-sequence generator based on Hugging Face's transformers. You can use it with any Hugging Face language model that extends GenerationMixin. See also How to Generate Text.

Usage

To initialize a Generator, run:

from haystack.nodes import RAGenerator

generator = 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, run:

from haystack.pipelines import GenerativeQAPipeline

pipeline = 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, run:

result = generator.predict(
    query='What are the best party games for adults?',
    documents=[doc1, doc2, doc3...],
    top_k=top_k
)