Datasets:

Modalities:
Text
Formats:
parquet
ArXiv:
Libraries:
Datasets
Dask
loubnabnl HF Staff commited on
Commit
86621d3
Β·
verified Β·
1 Parent(s): 23574ee

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +59 -0
README.md CHANGED
@@ -54,3 +54,62 @@ configs:
54
  - split: train
55
  path: kaggle/train-*
56
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
54
  - split: train
55
  path: kaggle/train-*
56
  ---
57
+
58
+ # πŸ“ GitHub Issues & Notebooks
59
+ ## Description
60
+ πŸ“ GitHub Issues & Notebooks is a collection of code datasets intended for language models training, they are sourced from GitHub issues, Kaggle notebooks, and Jupyter notebooks. These datasets are part of the [StarCoder2](https://arxiv.org/abs/2402.19173) model training corpus and a modified subset of [bigcode/StarCoder2-Extras](https://huggingface.co/datasets/bigcode/starcoder2data-extras) dataset. We reformat the samples to remove StarCoder2's special tokens and use natural text to delimit comments in issues and display kaggle notebooks in markdown and code blocks.
61
+
62
+ The dataset includes three subsets:
63
+
64
+ - πŸ› GitHub Issues – 11B tokens of technical discussions and issue tracking from GitHub repositories.
65
+ - πŸ“Š Kaggle Notebooks – 2B tokens of data analysis notebooks curated from Kaggle.
66
+ - πŸ’» Jupyter Notebooks – 16B tokens of Jupyter notebooks converted to Python scripts for better processing.
67
+ These subsets have undergone filtering to remove low-quality content, duplicates, more details in StarCoder2 [paper](https://arxiv.org/abs/2402.19173)
68
+
69
+ ## How to load the dataset
70
+
71
+ You can load a specific subset using the following code:
72
+
73
+ ```python
74
+ from datasets import load_dataset
75
+
76
+ data = load_dataset("HuggingFaceTB/github-issues-notebooks", "issues", split="train") # GitHub Issues
77
+ data = load_dataset("HuggingFaceTB/github-issues-notebooks", "kaggle", split="train") # Kaggle Notebooks
78
+ data = load_dataset("HuggingFaceTB/github-issues-notebooks", "jupyter", split="train") # Jupyter Notebooks
79
+ ```
80
+
81
+ ## Dataset curation
82
+ These curation details are from the SterCoder2 pipeline. The original datasets can be found at: https://huggingface.co/datasets/bigcode/starcoder2data-extras
83
+
84
+ ### πŸ› GitHub Issues
85
+ The GitHub Issues dataset consists of discussions from GitHub repositories, sourced from GHArchive. It contains issue reports, bug tracking, and technical Q&A discussions.
86
+
87
+ To ensure high-quality data, the StarCoder2 processing pipeline included:
88
+
89
+ - Removing bot-generated comments and auto-replies from email responses.
90
+ - Filtering out short issues (<200 characters) and extremely long comments.
91
+ - Keeping only discussions with multiple users (or highly detailed single-user reports).
92
+ - Anonymizing usernames while preserving the conversation structure.
93
+ - This cleaning process removed 38% of issues, ensuring a high-quality dataset with technical depth.
94
+ - More details can be found in the StarCoder2 paper.
95
+
96
+ ## πŸ’» Jupyter Notebooks
97
+ The Jupyter Notebooks dataset consists of 4M deduplicatedstructured notebooks, converted to Python scripts using Jupytext.
98
+
99
+ ## πŸ“Š Kaggle Notebooks
100
+ The Kaggle Notebooks are sourced from the [Meta Kaggle Code](https://www.kaggle.com/datasets/kaggle/meta-kaggle-code) dataset. They were cleaned using a multi-step filtering process, which included:
101
+
102
+ - Removing notebooks with syntax errors or less than 100 characters.
103
+ - Extracting metadata for notebooks that reference Kaggle datasets, when possible we retrieve the datasets and append information about the data to the beginning of the notebooks (description, `ds.info()`output and 4 examples)
104
+ - Filtering out duplicates, which reduced the dataset volume by 78%
105
+
106
+
107
+ ## Citation
108
+ ```
109
+ @article{lozhkov2024starcoder,
110
+ title={Starcoder 2 and the stack v2: The next generation},
111
+ author={Lozhkov, Anton and Li, Raymond and Allal, Loubna Ben and Cassano, Federico and Lamy-Poirier, Joel and Tazi, Nouamane and Tang, Ao and Pykhtar, Dmytro and Liu, Jiawei and Wei, Yuxiang and others},
112
+ journal={arXiv preprint arXiv:2402.19173},
113
+ year={2024}
114
+ }
115
+ ```