API Reference

QA Pipelines#

The pipeline allows the components to be used as building blocks and allows switching out alternative implementations for the retriever and reader.

We show some pipeline examples for the Open Retrieval Question Answering. Open retrieval systems query large document stores for relevant passages. These passages are used by the reader to produce the answer.

Generative QA with LLM#

We prompt pre-trained large language models (LLMs) to answer questions conditioned on the retrieved passages. We support integration with the OpenAI’s ChatGPT(gpt-3.5-turbo) and InstructGPT(text-davinci-003), using an OpenAI API key (https://platform.openai.com/account/api-keys). Examples can be found in the notebook.

We also support the Flan-T5 LM and examples can be found in the notebook.

  • Step 1: Initialize the retriever.

retriever = ColBERTRetriever(index_root = index_root, index_name = index_name, collection = collection, max_num_documents = 3)
retriever.load()

  • Step 2: Initialize the reader model.

We can use a reader based on ChatGPT.

reader = PromptGPTReader(api_key='API KEY HERE', model_name="gpt-3.5-turbo")
reader.load()

Alternatively, we can use a reader based on Flan-T5.

reader = PromptFLANT5Reader(model_name="google/flan-t5-xxl")
reader.load()

  • Step 3: Initialize the QA pipeline.

llm_pipeline = QAPipeline(retriever, reader)

  • Step 4: Execute the QA pipeline in inference mode.

queries=["What causes the trail behind jets at high altitude?"]
prompt_prefix = "Answer the following question after looking at the text."
answers = llm_pipeline.run(query, prefix=prompt_prefix)

Open Retrieval with Extractive Reader#

The pipeline uses a ColBERT retriever to obtain supporting passages. The reader extracts the best answers from the passages and provides evidence for its predictions.

  • Step 1: Initialize the retriever.

retriever = ColBERTRetriever(index_root = index_root, index_name = index_name, collection = collection, max_num_documents = 3)
retriever.load()

  • Step 2: Initialize the reader model.

reader = ExtractiveReader(model_name_or_path = model)
reader.load()

  • Step 3: Initialize the QA pipeline.

extractive_pipeline = QAPipeline(retriever, reader)

  • Step 4: Execute the pipeline in inference mode.

queries=["Which country is Canberra located in?"]
answers = pipeline.run(query)

Long Form Question Answering (LFQA)#

LFQA is a generative task where the retrieved passages are used to generate a complex multi-sentence answer.

In this example we show a QA Pipeline using a ColBERT retriever and a Fusion in Decoder (FID) generator.

Instructions to create a ColBERT index and an FiD model for KILT-ELI5 can be found here

  • Step 1: Initialize the retriever.

retriever = ColBERTRetriever(index_root = index_root, index_name = index_name, collection = collection, max_num_documents = 3)
retriever.load()

  • Step 2: Initialize the reader model.

reader = GenerativeFiDReader(model_name_or_path = model)
reader.load()

  • Step 3: Initialize the QA pipeline.

lfqa_pipeline = QAPipeline(retriever, reader)

  • Step 4: Execute the LFQA pipeline in inference mode.

queries=["What causes the trail behind jets at high altitude?"]
answers = lfqa_pipeline.run(query)