---
language:
- fa
size_categories:
- 1K<n<10K
task_categories:
- question-answering
- text-generation
- text2text-generation
pretty_name: PersianQA Instructions
dataset_info:
  features:
  - name: input
    dtype: string
  - name: output
    dtype: string
  splits:
  - name: train
    num_bytes: 18504268
    num_examples: 9008
  - name: validation
    num_bytes: 1882042
    num_examples: 930
  download_size: 3087085
  dataset_size: 20386310
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
tags:
- PersianQA
- Instruction Finetuning
---
# [PersianQA Instructions](https://huggingface.co/datasets/ErfanMoosaviMonazzah/persian_qa_instructions)
This is a pre-processed version of [PersianQA](https://huggingface.co/datasets/SajjadAyoubi/persian_qa) ready to be used for instruction-finetuning of LLMs for persian instruction following.
Corrections:
- There were some misusage of character '\u200c' (half-space): All the instances of '\u200c' were replaced with normal space character ' '.
- Columns were renamed to follow alpaca format:
  - question --> instruction
  - context --> input
  - answers --> output
- Columns 'id' & 'title' were removed.
- For the output, the original [PersianQA](https://huggingface.co/datasets/SajjadAyoubi/persian_qa) provided more than one answers in some cases. We selected the longest answer in that case since the LLMs are comfortable with generating long texts during inference and having longer phrases and sentences might be better during the instruction-finetuning.
- There are some cases where the answer (output) is not presented in the context (input), in that case the output is:
  - Persian: 'با عرض پوزش، در متن داده شده نمی توانم پاسخ سوال شما را پیدا کنم.'
  - English Translation: 'Sorry, I cannot find the answer to your question in the given text.'