File size: 1,517 Bytes
1d5eb18
 
 
 
 
 
 
 
 
 
 
14004e6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1d5eb18
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
---
license: odc-by
task_categories:
- text-classification
- token-classification
- question-answering
- text-generation
- text2text-generation
size_categories:
- 100K<n<1M
---
# Essential Web v1.0 - 1M Token Sample

Approximately 1,000,000 tokens sampled from Essential Web v1.0.

## Dataset Info

- **Target**: 1,000,000 tokens  
- **Actual**: ~1,099,800 tokens (estimated)
- **Source**: [EssentialAI/essential-web-v1.0](https://huggingface.co/datasets/EssentialAI/essential-web-v1.0)

## Schema

This sample preserves ALL columns from the original dataset, including:
- `id`: Document ID
- `text`: Text content
- `metadata`: URL and source information
- `quality_signals`: RedPajama quality metrics
- `eai_taxonomy`: Essential AI taxonomy labels  
- `pid`: Partition ID
- And all other original columns

## Usage

```python
from datasets import load_dataset

dataset = load_dataset("sumuks/essential-web-v1.0-sample-1M")

# Access the data with all columns
example = dataset['train'][0]
print(example['text'][:200] + "...")

# Access quality signals
print(example['quality_signals'])

# Access taxonomy
print(example['eai_taxonomy'])
```

## File Structure

The dataset is split across multiple parquet files in the `data/` directory:
- `data/part-00000.parquet`
- `data/part-00001.parquet`  
- etc.

HuggingFace datasets automatically loads all parts as a single dataset.

## Sampling Method

- Random sampling across snapshots
- Preserves all original columns and metadata
- Token estimation: ~600 tokens per row