README.md

---
dataset_info:
  features:
  - name: function_name
    dtype: string
  - name: docstring
    dtype: string
  - name: masked_code
    dtype: string
  - name: implementation
    dtype: string
  - name: start_line
    dtype: int32
  - name: end_line
    dtype: int32
  - name: file_content
    dtype: string
  splits:
  - name: train
    num_bytes: 420616564
    num_examples: 2760
  download_size: 64655948
  dataset_size: 420616564
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---


# Stack-Smol-Docstrings

This dataset contains Python functions extracted from [the-stack-smol](https://huggingface.co/datasets/bigcode/the-stack-smol), 
filtered for high-quality docstrings and implementations. Each sample includes 
the function's docstring, implementation, and a masked version of the code where the 
function is replaced with a comment.

The dataset is designed for code completion tasks where a model needs to restore a 
function that has been replaced with a comment. The model is provided with:
1. The full file context with the function replaced by a comment
2. The docstring of the function
3. The function name

The model's task is to generate code that replaces the comment with a proper implementation
of the function based on the docstring and surrounding context.

## Dataset Structure
Each sample contains:
- `function_name`: Name of the function
- `docstring`: The function's docstring
- `masked_code`: The full file with the function replaced by a comment
- `implementation`: The original function implementation
- `start_line`: The starting line number of the function in the original file
- `end_line`: The ending line number of the function in the original file
- `file_content`: The full original file content

## Quality Filtering
Functions are filtered based on:
- Docstring quality (length, structure, descriptiveness)
- Implementation quality (no SQL strings, reasonable number of variables, sufficient complexity)