Update README.md

README.md

@@ -6563,3 +6563,61 @@ configs:
   - split: test
     path: data/test-*
 ---
+
+# 📄 Paper2Code: Automating Code Generation from Scientific Papers in Machine Learning
+- **Repository:** [https://github.com/going-doer/Paper2Code]
+- **Paper** [https://arxiv.org/abs/2504.17192]
+
+## Paper2Code Benchmark
+
+### Dataset Description
+
+The Paper2Code Benchmark is designed to evaluate the ability to reproduce methods and experiments described in scientific papers.<br/>
+We collected 90 papers from ICML 2024, NeurIPS 2024, and ICLR 2024, selecting only those with publicly available GitHub repositories.<br/>
+To ensure manageable complexity, we filtered for repositories with fewer than 70,000 tokens.
+
+Using a model-based evaluation, we selected the top 30 papers from each conference based on repository quality.
+
+For more details, refer to Section 4.1 "Paper2Code Benchmark" of the paper.
+
+
+## Uses
+
+<!-- Address questions around how the dataset is intended to be used. -->
+```python
+from datasets import load_dataset
+
+dataset = load_dataset("iaminju/paper2code", split="test")
+```
+
+For access to the benchmark files (including pdf files), please refer to the [**Paper2Code data directory**](https://github.com/going-doer/Paper2Code/tree/main/data/paper2code) in our GitHub repository.
+
+## Dataset Structure
+
+<!-- This section provides a description of the dataset fields, and additional information about the dataset structure such as criteria used to create the splits, relationships between data points, etc. -->
+```
+Dataset({
+    features: ['paper', 'source', 'repo_name', 'repo_url', 'paper_json', 'paper_cleaned_json', 'conference'],
+    num_rows: 90
+})
+```
+- `paper`: Title of the paper.  
+- `source`: Presentation type — oral or poster.  
+- `repo_name`: Name of the repository provided by the original authors.  
+- `repo_url`: URL of the repository provided by the original authors.  
+- `paper_json`: Parsed JSON version of the paper. We use [s2orc-doc2json](https://github.com/allenai/s2orc-doc2json) for this conversion.  
+- `paper_cleaned_json`: Preprocessed version of the paper used by PaperCoder.  
+- `conference`: The conference where the paper was accepted - icml2024, iclr2024 or nips2024
+
+
+## Citation
+
+<!-- If there is a paper or blog post introducing the dataset, the APA and Bibtex information for that should go in this section. -->
+```
+@article{seo2025paper2code,
+  title={Paper2Code: Automating Code Generation from Scientific Papers in Machine Learning},
+  author={Seo, Minju and Baek, Jinheon and Lee, Seongyun and Hwang, Sung Ju},
+  year={2025},
+  url={https://arxiv.org/pdf/2504.17192}
+}
+```