README.md

metadata

dataset_info:
  features:
    - name: anchor
      dtype: string
    - name: positive
      dtype: string
    - name: negative
      dtype: string
  splits:
    - name: train
      num_bytes: 143845823
      num_examples: 317521
    - name: test
      num_bytes: 61833745
      num_examples: 136081
  download_size: 51750220
  dataset_size: 205679568
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*

Description

This dataset has been built from the MBPP dataset for fine tuning dense retrieval models. The dataset was created by using the first 70% points from the MBPP dataset. We created triplets corresponding to all negatives for a positive pair. Hence there are n * (n - 1) triplets for n pairs(since we have n-1 negative examples for every anchor-positive pair). Using a random seed of 10, we split these triplets into train and test subsest with a 70:30 ratio.

Fields

1. anchor - The question corresponding to a code snippet
2. positive - The ground truth answer for the corresponding question
3. negative - Any other code snippet from the dataset not corresponding to the question