Dataset Viewer
Auto-converted to Parquet Duplicate
The dataset viewer is not available for this split.
Parquet error: Scan size limit exceeded: attempted to read 725442754 bytes, limit is 300000000 bytes Make sure that 1. the Parquet files contain a page index to enable random access without loading entire row groups2. otherwise use smaller row-group sizes when serializing the Parquet files
Error code:   TooBigContentError

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

llm-und/pretrain-pt-es-hi

Built from data/processed/{pt,es,hi}/*.parquet, produced by scripts/run_dedup_datatrove.py (MinHash dedup across pt-fineweb2, es-fineweb2, hi-fineweb2, and hi-sangraha -- see README.md). Doc counts below reflect whatever run produced the input.

config docs
pt 195,865,727
es 434,552,847
hi 36,430,466

Load a single language:

from datasets import load_dataset
ds = load_dataset("{repo_id}", "pt")  # or "es", "hi"

Load any combination (e.g. all three) by concatenating -- this is a one-line operation at load time, not a separate config baked into the dataset:

from datasets import load_dataset, concatenate_datasets
ds = concatenate_datasets([load_dataset("{repo_id}", lang, split="train") for lang in ["pt", "es", "hi"]])
Downloads last month
1,362