sumanthd/

c3po-mixed-pretraining-corpus

Mixed pretraining corpus sampled from seven public Hugging Face datasets, filtered by token length (no truncation). Built by the c3po project.

Documents: 25,000 Tokenizer: google/gemma-4-26b-a4b-it Token window: 1,000 – 32,000 Built: 2026-05-23T12:11:43.084386+00:00

	Sources




	

Source Documents HF repo

Clean Wikipedia 4,940 DragonLLM/Clean-Wikipedia-English-Articles

Cosmopedia v2 2,870… See the full description on the dataset page: https://huggingface.co/datasets/sumanthd/c3po-mixed-pretraining-corpus.