README.md

metadata

task_categories:
  - text-generation
language:
  - en
size_categories:
  - 1B<n<10B

Dataset: LLaDA-Sample-10BT
Base: HuggingFaceFW/fineweb (subset sample-10BT)
Purpose: Training LLaDA (Large Language Diffusion Models)

Preprocessing

• Tokenizer: GSAI-ML/LLaDA-8B-Instruct
• Chunking: Up to 4,096 tokens per chunk (1% of chunks randomly sized between 1–4,096 tokens)
• Noisy masking: Applied with noise factor ε = 1×10⁻³
• Fields per chunk (PyTorch tensors):
  • input_ids
  • noisy_input_ids
  • mask
  • t (time scalar)

Statistics

• Total chunks: ~2,520,000
• Shards: 252 .pt files
• Chunks per file: 10,000
• Average file size: ~702–708 MB
• Total size: ~166 GB

Usage

This dataset is used for training in the LLaDA-from-scratch GitHub repository, where you’ll find the full data pipeline and training scripts.