Ujjwal-Tyagi/moshub

Mos.Hub Code Dataset

A comprehensive code dataset compiled from Mos.Hub, Moscow's official code hosting platform operated by the Moscow Government. This dataset is designed to support training code models with authentic Russian development practices and documentation.

---

Overview

The Mos.Hub Code Dataset represents a significant code corpus from Russia's governmental and municipal code hosting platform, capturing diverse projects across 297 programming languages. It serves as a valuable resource for developing multilingual code understanding models reflecting Russian coding practices and standards.

Key Statistics

Metric	Value
Total Files	15,740,580
Total Repositories	16,130
Compressed Size	529 MB (Parquet)
Uncompressed Size	~29 GB
Programming Languages	297
File Format	Single Parquet file

---

Dataset Characteristics

Scope and Coverage

This dataset captures code from over 16,000 repositories hosted on Mos.Hub, including:

• Russian government and public sector code: Repositories from Moscow Government and related organizations, featuring Russian comments and documentation
• Diverse language ecosystem: Support for 297 distinct programming languages
• Wide project range: From municipal services to development projects and tools
• Quality-assured: Binary files and low-quality content have been systematically removed

Programming Languages

The dataset encompasses 297 languages. The 30 most represented languages by file count are:

Rank	Language	File Count
1	Ruby	8,333,731
2	JavaScript	1,786,730
3	YAML	1,757,614
4	Vue	699,171
5	Markdown	639,585
6	Haml	538,837
7	GraphQL	269,485
8	JSON	214,354
9	PHP	191,150
10	SVG	172,884
11	Shell	172,451
12	Go	88,089
13	Ignore List	87,432
14	SCSS	80,716
15	Python	77,532
16	C++	63,177
17	HTML+ERB	62,605
18	Text	48,400
19	Jest Snapshot	43,638
20	HTML	42,489
21	C	38,354
22	reStructuredText	26,342
23	Rust	24,818
24	E-mail	23,993
25	XML	22,715
26	Java	14,807
27	Gettext Catalog	14,429
28	C#	13,405
29	CSS	12,657
30	Protocol Buffer Text Format	12,181

---

Dataset Structure

Data Fields

Each record contains three fields providing content and metadata:

Field	Type	Description
file_text	string	Complete file content in UTF-8 encoding
language	string	Programming language identified using github-linguist
file_name	string	Name of the source file

Sample Record

{
    "file_text": "package main\n\nimport \"fmt\"\n\nfunc main() {\n    fmt.Println(\"Hello\")\n}\n",
    "language": "Go",
    "file_name": "main.go"
}

File Format

• Format: Apache Parquet
• Structure: Single consolidated file (data.parquet)
• Encoding: UTF-8
• Split: All examples are included in a single training split (no validation or test splits)

---

Data Creation Process

Language Detection Methodology

Programming languages are identified using github-linguist, GitHub's robust library for language detection. This ensures consistent and reliable classification across all files in the dataset.

Source Data

All data originates from public repositories hosted on Mos.Hub, Moscow's official code hosting platform managed by the Moscow Government.

Quality Filtering

The dataset has undergone systematic filtering to ensure quality and usability:

Deduplication

• Files have been deduplicated to ensure each code file appears only once in the dataset

Binary File Removal

• Binary files have been systematically excluded from the dataset
• Only text-based source code files are retained

Text Encoding Validation

• Files must be valid UTF-8 encoded text to be included
• Non-text and improperly encoded files are excluded

---

Usage Considerations

Data Privacy and Security

The dataset may contain sensitive information that requires careful handling:

• Email Addresses: Present in code comments, documentation, or configuration files
• Credentials: Accidentally committed API keys or authentication tokens
• Personal Information: Names, phone numbers, and other identifiable data in comments or documentation

Users should implement appropriate filtering and anonymization when preparing data for model training.

Licensing and Attribution

This dataset has been compiled with careful consideration of the licenses used in source repositories. Any use of code or data derived from this dataset should be done responsibly and ethically.

Users are responsible for:

• Respecting the intellectual property rights of original authors
• Using the data responsibly and ethically
• Understanding any applicable license restrictions
• Providing appropriate attribution when required

Responsible Use

The dataset should be used with awareness of:

• The public sector origin of many repositories
• The importance of code security and data protection
• Compliance with Russian data protection regulations
• Ethical considerations in code model training and deployment

---

Technical Details

Source: Public repositories hosted on Mos.Hub

Annotations: Machine-generated (language detection)

Multilingual Support: Includes multilingual code and documentation with emphasis on Russian content

Task Categories: Text generation, code modeling, language understanding

Tags: Code, Russian language, multilingual, public sector development

---