all-wikis-mix
FineWeb-style Parquet shards created by merging all the already-built monolingual Wikipedia datasets into a single dataset.
Why this exists
Training pipelines that round-robin by shard can “loop” on tiny datasets (repeating the same few shards). A single merged dataset has a healthier shard count and makes mixing more stable.
What’s inside
- Format:
nanochat-parquet-v1 - Layout:
shard_*.parquet+metadata.json - Text column:
text - Parquet settings: zstd (level 3),
row_group_size=1024,use_dictionary=False,write_statistics=False
Loading
from datasets import load_dataset
ds = load_dataset("JayJayThrowThrow/all-wikis-mix", split="train")
print(ds.column_names) # ["text"]