Question Answering when context is provided

[NeuroinformaticaFBF]

Hello

I just trained a QA model on my dataset by using Haystack libraries.
Now I would like to train it by providing it a context and a question. There's no need of performing a document retrieval.

I saw a lot of tutorial for complex QA pipelines, but I wasn't able to find how to perform this simple task (at least, I assume it's simple).

Can someone help, or point some kind of tutorial/guide?

Many thanks

[julian-risch]

For question answering on a provided context, without retrieval, all you need is a Reader component.

from haystack.nodes import FARMReader
reader = FARMReader(
            model_name_or_path="deepset/bert-medium-squad2-distilled",
            return_no_answer=True,
        )
prediction = reader.predict(query="Who lives in Berlin?", documents=docs, top_k=5)

[julian-risch]

@NeuroinformaticaFBF You can also wrap the reader in a pipeline if you want:

from haystack.pipelines.base import Pipeline
pipeline = Pipeline()
pipeline.add_node(component=reader, name="Reader", inputs=["Query"])

[julian-risch]

As you see, you don't need to train a separate model. The Reader that is used in a QA pipeline with retrieval just works the same way without retrieval.

[NeuroinformaticaFBF]

Thanks for the reply, I can use it now.
One more (quick) question: the f1-score calculated during the fine-tuning of a QAbot (f1_text_answer), is it Macro-f1 or Weighted-f1?

Thanks!