Hugging Face's logo

Datasets:
LevelUp2x
/
nova-corpus

Modalities:
Tabular
Text
Formats:
json
Size:
< 1K
Libraries:
Datasets
Dask
Dataset card Data Studio Files
xet
Community
1
nova-corpus / documentation /README.md
ADAPT-Chase's picture
ADAPT-Chase
Upload documentation/README.md with huggingface_hub
6a6070f verified
preview code
|
raw
history blame contribute delete
5.41 kB

Corpus Data Directory

Location

/data/adaptai/corpus-data

Purpose

This directory serves as the central storage location for all corpus data pulled from Nebius S3 and other sources. It is the primary input directory for the bleeding-edge ETL pipeline.

Data Organization

Directory Structure 

/data/adaptai/corpus-data/
├── nebius-oscar/                 # OSCAR corpus from Nebius S3
│   ├── unsharded/               # Unsharded multilingual data
│   ├── wikipedia/               # Wikipedia dumps
│   └── commoncrawl/             # Common Crawl data
├── mounted-s3/                  # Symlinks to mounted S3 buckets
│   ├── oscar-corpus -> /mnt/s3/oscar-corpus
│   └── other-buckets/           # Additional S3 buckets
├── processed/                   # Processed data ready for analysis
│   ├── flowetl-transformed/    # FlowETL processed files
│   └── cleaned/                # Cleaned and normalized data
└── backups/                     # Corpus data backups
    └── YYYY-MM-DD/             # Date-based backup folders

Data Sources

Primary Sources

  1. Nebius S3 - OSCAR Corpus

    • Open Super-large Crawled Aggregated coRpus
    • 100+ languages
    • Petabyte-scale multilingual data
    • Real-time streaming capability

  2. Wikipedia Dumps

    • Multilingual Wikipedia articles
    • Structured text data
    • Regular updates

  3. Common Crawl

    • Web crawl data
    • Diverse content types
    • Massive scale

Integration Methods

Direct Mount (Recommended) 

# Mount Nebius S3 buckets
s3fs oscar-corpus /mnt/s3/oscar-corpus -o url=https://storage.yandexcloud.net

# Access data through symlinks
ls -la /data/adaptai/corpus-data/mounted-s3/

Automated Pull Script 

# Use the automated puller
python3 /data/adaptai/bleeding-edge-etl/nebius_s3_mount.py

# Environment variables required:
export Nebius_ACCESS_KEY=your_access_key
export Nebius_SECRET_KEY=your_secret_key

Manual Download 

# For specific file downloads
aws s3 sync s3://oscar-corpus/unsharded/ /data/adaptai/corpus-data/nebius-oscar/unsharded/

Processing Pipeline

FlowETL Integration

Corpus data in this directory is automatically processed by:

  1. FlowETL - Autonomous transformations
  2. Apache NiFi - Orchestration and flow management
  3. Apache Drill - Schema-free querying
  4. CWB/ANNIS - Linguistic analysis

Data Flow 

Nebius S3 → /data/adaptai/corpus-data/ → FlowETL → Processed Data → Analysis

Storage Requirements

Capacity

  • Initial: 10TB+ for sample datasets
  • Production: 100TB+ for full corpus processing
  • Scalable: Designed for petabyte-scale growth

Performance

  • SSD Storage: /data partition optimized for high I/O
  • Memory Caching: DragonFly/Redis for frequent access
  • Network: High-throughput connectivity to Nebius S3

Access Patterns

Read Access

  • FlowETL transformation engine
  • Apache Drill for SQL queries
  • CWB/ANNIS for linguistic analysis
  • Research and development tools

Write Access

  • Automated S3 sync processes
  • Manual data ingestion
  • Processing pipeline outputs
  • Backup and archival systems

Backup Strategy

Automated Backups 

# Daily incremental backups
rsync -av --delete /data/adaptai/corpus-data/ /backup/corpus-data/daily/

# Weekly full backups
tar -czf /backup/corpus-data/weekly/$(date +%Y-%m-%d).tar.gz /data/adaptai/corpus-data/

Cloud Backup

  • Regular sync to Nebius S3 for disaster recovery
  • Versioned backups for data recovery
  • Geographic redundancy

Security

Access Control

  • Role-based permissions
  • Audit logging
  • Encryption at rest and in transit

Data Protection

  • Anonymization where required
  • Compliance with usage agreements
  • Regular security audits

Monitoring

Health Checks 

# Disk space monitoring
df -h /data/adaptai/corpus-data

# Data integrity checks
find /data/adaptai/corpus-data -name "*.jsonl" -exec jsonschema -i {} \;

# Access monitoring
inotifywait -m -r /data/adaptai/corpus-data

Performance Metrics

  • Throughput: GB/s processed
  • Latency: End-to-end processing time
  • Quality: Data validation results
  • Utilization: Storage capacity metrics

Troubleshooting

Common Issues

  1. Permission Denied

    sudo chown -R $(whoami):$(whoami) /data/adaptai/corpus-data

  2. Disk Space Full

    # Clean up temporary files
find /data/adaptai/corpus-data -name "*.tmp" -delete

  3. S3 Mount Failed

    # Check credentials
cat /etc/passwd-s3fs

# Remount
sudo umount /mnt/s3/*
sudo s3fs oscar-corpus /mnt/s3/oscar-corpus -o url=https://storage.yandexcloud.net

Related Components

ETL Pipeline

  • FlowETL: /data/adaptai/bleeding-edge-etl/flowetl/
  • Apache NiFi: /data/adaptai/bleeding-edge-etl/nifi/
  • Apache Drill: /data/adaptai/bleeding-edge-etl/drill/
  • CWB/ANNIS: /data/adaptai/bleeding-edge-etl/corpus-analysis/

Infrastructure

  • Nebius S3: Cloud object storage
  • DragonFly: High-performance cache
  • Redis: Traditional caching
  • Qdrant: Vector database for analysis

Maintained by: ETL Team - Bleeding-Edge Corpus Aggregation Last Updated: August 24, 2025 Status: ACTIVE - Ready for Data Ingestion