---
language:
- en
---

# Dataset description

Number of articles for the English wiki:
- 2014: 4.599.592
- 2016: 5.144.403
- 2018: 5.599.764
- 2020: 6.037.287
- 2022: 6.291.973
- 2024: 6.629.861
    
with 8k documents per gzipped json, we get approx. 800 files per year, so aprrox. 3500 files in total. Approx. 190-200 MB per bin of 8k examples, 63 in gzipped mode.

# Dataset sizes
There are different splits and dataset sizes, which are all subsets of the `full` set.

| Config   | Train | Validation | Estimate |
|----------|-------|------------|----------|
| tiny     | 16k docs (2 shards)     | 8k docs (1 shard)          | 0.1GB    |
| small    | 800k docs (100 shards)   | 16k docs (2 shards)          | 4GB      |
| medium   | 6M docs (750 shards)   | 16k docs (2 shards)          | 30GB     |
| large    | 12M docs (1500 shards)  | 24k docs (3 shards)          | 59GB     |
| full     | 28M docs (3497 shards)  | 32k docs (4 shards)          | 137GB    |