icfoss/malayalam-fineweb-edu-mt

Malayalam FineWeb-Edu MT

Malayalam FineWeb-Edu MT is a large-scale monolingual Malayalam corpus created by machine-translating educational web text from FineWeb-Edu into Malayalam. At release time, to the maintainers' knowledge, this is the largest publicly released monolingual Malayalam corpus prepared for language-model pretraining and research.

The corpus is distributed as Apache Arrow-compatible Parquet files, with one Malayalam text sample per row.

Dataset Details

• Language: Malayalam (ml)
• Format: Parquet
• Split: train
• Rows: derived from three local source files: mal1.txt, mal2.txt, and mal3.txt
• Text source: machine translation of FineWeb-Edu educational web content
• Intended use: Malayalam language-model pretraining, continued pretraining, tokenizer training, filtering research, corpus analysis, and other non-commercial or commercial research workflows permitted by the license

Columns

Column	Type	Description
id	string	Stable row identifier built from the source file name and line number.
text	string	Malayalam text.
language	string	Always ml.
source_file	string	Original local shard name: mal1, mal2, or mal3.
line_number	int64	1-based line number in the original source file.
machine_translated	bool	Always true.

Loading

from datasets import load_dataset

ds = load_dataset("icfoss/malayalam-fineweb-edu-mt", split="train")
print(ds[0]["text"])

For streaming:

from datasets import load_dataset

ds = load_dataset("icfoss/malayalam-fineweb-edu-mt", split="train", streaming=True)
for row in ds.take(3):
    print(row["text"])

Provenance

This dataset is a Malayalam machine-translated derivative of FineWeb-Edu. FineWeb-Edu is an educational subset of FineWeb, filtered using an educational quality classifier. The original FineWeb-Edu dataset is released by Hugging Face under the Open Data Commons Attribution License.

Original dataset: HuggingFaceFW/fineweb-edu

Quality Notes

The corpus is machine translated and should be treated as synthetic Malayalam text. It may contain translation artifacts, named entities from global web pages, copied markup fragments, malformed punctuation, duplicated content, or occasional non-Malayalam spans inherited from the source material or translation process. Users training models on this dataset may want to apply additional deduplication, quality filtering, toxicity filtering, PII filtering, and domain-specific validation.

License

This dataset is released under the Open Data Commons Attribution License (odc-by), following the license of FineWeb-Edu. Use of the underlying source material may also be subject to Common Crawl's Terms of Use and any applicable rights in the original web pages.

Citation

If you use this dataset, please cite this dataset and the original FineWeb-Edu work:

@misc{malayalam_fineweb_edu_mt,
  title = {Malayalam FineWeb-Edu MT},
  author = {ICFOSS},
  year = {2026},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/icfoss/malayalam-fineweb-edu-mt}
}

@misc{lozhkov2024finewebedu,
  title = {FineWeb-Edu: the Finest Collection of Educational Content},
  author = {Lozhkov, Anton and Ben Allal, Loubna and von Werra, Leandro and Wolf, Thomas},
  year = {2024},
  url = {https://huggingface.co/datasets/HuggingFaceFW/fineweb-edu},
  doi = {10.57967/hf/2497},
  publisher = {Hugging Face}
}