File size: 2,328 Bytes
ccb197a e647a63 ccb197a e647a63 ccb197a 5df15d9 9d2c78f 5df15d9 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 |
---
dataset_info:
features:
- name: token_count
dtype: int32
- name: text
dtype: string
- name: score
dtype: float64
splits:
- name: train
num_bytes: 10766525501.0
num_examples: 2000000
download_size: 5790667341
dataset_size: 10766525501.0
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# DraftWeb-v1 Dataset Card
## Dataset Description
A large-scale dataset created by combining educational mathematics problems and web content to create a continued pretraining dataset for LLMs. The dataset contains 2,000,000 high-quality examples sorted by content relevance scores.
## Data Source
- **Math Content**: Hugging Face `finemath-4plus` dataset (educational math problems)
- **Web Content**: Hugging Face `fineweb-edu` dataset (web pages from Common Crawl)
## Training Data
- 1,000,000 math examples with token counts
- 1,000,000 web examples with token counts
- Combined and shuffled for model training
## Dataset Statistics
- **Total Examples**: 2,000,000
- **Total Tokens**: 2.7 billion
- Math Token Count: 1.61 billion
- Web Token Count: 1.11 billion
- **Languages**: English (primary)
- **Content Types**: Text-based math problems and web pages
## Data Processing
1. Removed non-essential metadata columns
2. Standardized token count measurement
3. Sorted by source dataset scores (descending)
4. Selected top 1 million examples from each source
5. Combined and shuffled for model training
## Usage
- Fine-tuning foundation language models
- Instruction tuning for educational applications
- Web content understanding tasks
- Research in educational AI systems
## Licensing
- Inherits licenses from source datasets:
- `finemath-4plus`: CC BY 4.0
- `fineweb-edu`: CC BY-SA 4.0
- For commercial use, please verify specific license requirements
## Evaluation
- Original datasets contained quality scores used for selection
- Recommended evaluation metrics:
- Text quality scoring
- Relevance prediction accuracy
- Downstream task performance (e.g., question answering)
## Dataset Provenance
- Hosted on Hugging Face Hub: `qingy2024/DraftWeb-v1`
- Creation date: July 14, 2025
- Version: 1.0
## Dataset Expansion
- Potential future versions:
- Multilingual expansion
- Specialized domain versions
- Time-based snapshots |