README.md

---
license: mit
dataset_info:
- config_name: documents
  features:
  - name: question_id
    dtype: int64
  - name: documents
    sequence: string
  splits:
  - name: train
    num_bytes: 18770828
    num_examples: 1172
  download_size: 10608695
  dataset_size: 18770828
- config_name: question-answer
  features:
  - name: id
    dtype: int64
  - name: question
    dtype: string
  - name: answer
    dtype: string
  - name: relevant_passage_ids
    dtype: string
  splits:
  - name: train
    num_bytes: 386574
    num_examples: 1172
  download_size: 252573
  dataset_size: 386574
configs:
- config_name: documents
  data_files:
  - split: train
    path: documents/train-*
- config_name: question-answer
  data_files:
  - split: train
    path: question-answer/train-*
task_categories:
- question-answering
language:
- en
tags:
- biology
size_categories:
- 1K<n<10K
---

This dataset is a fork of [rag-datasets/rag-mini-bioasq](https://huggingface.co/datasets/rag-datasets/rag-mini-bioasq).

The use case for this dataset is to test and evaluate an LLM or a RAG application's ability to filter out irrelevant information to answer a given question.

This forks filters away all questions with one or more "NaN" passages, and adds an average of 80% irrelevant passages to each question.