metadata

dataset_info:
  features:
    - name: repo
      dtype: string
    - name: docfile_name
      dtype: string
    - name: doc_type
      dtype: string
    - name: intent
      dtype: string
    - name: license
      dtype: string
    - name: path_to_docfile
      dtype: string
    - name: relevant_code_files
      sequence: string
    - name: relevant_code_dir
      dtype: string
    - name: target_text
      dtype: string
    - name: relevant_code_context
      dtype: string
  splits:
    - name: test
      num_bytes: 227163668
      num_examples: 216
  download_size: 30375843
  dataset_size: 227163668
configs:
  - config_name: default
    data_files:
      - split: test
        path: data/test-*

🏟️ Long Code Arena (Module Summarization)

This is the data for Module Summarization benchmark as part of Long Code Arena provided by Jetbrains Research.

How-to

Load the data via load_dataset:

```
from datasets import load_dataset

dataset = load_dataset("JetBrains-Research/lca-module-to-text")
```

Datapoint Structure

Each example has the following fields:

Field	Description
`repo`	Name of the repository
`target_text`	Text of the target documentation file)
`docfile_name`	Name of the file with target documentation
`intent`	One sentence description of what expected in the documentation
`license`	License of the target repository
`relevant_code_files`	Pathes to relevant code files (files that are mentioned in target documentation)
`relevant_code_dir`	Pathes to relevant code directories (directory that are mentioned in target documentation)
`path_to_docfile`	Path to file with documentation (path to the documentation file in source repository)
`relevant_code_context`	Relevant code context collected from relevant code files and directories

Note: you may collect and use your own relevant context. Our context may not be suitable. Folder with zipped repositories can be found in Files and versions

Licences

We extracted repositories with permissive license (we used the most popular permissive licenses --- MIT, Apache-2.0, BSD-3-Clause, and BSD-2-Clause). The data points could be removed upon request

Metric

To compare predicted and ground truth metrics we introduce the new metric based on LLM as an assessor. Our approach involves feeding the LLM with relevant code and two versions of documentation: the ground truth and the model-generated text. The LLM evaluates which documentation better explains and fits the code. To mitigate variance and potential ordering effects in model responses, we calculate the probability that the generated documentation is superior by averaging the results of two queries:

For more details about metric implementation go to our github repository