vuhaian/

24_collected

23,000 text samples (1000 each) collected across 8 public HF datasets / 23 subset-configs, merged and shuffled into a single dataset. Built for continue-pretraining experiments against a diverse mix (general web, math, QA, code, reasoning) rather than a single source.

	Sources (1000 samples each, 23 buckets total)




	

source repo_id subset/config

automathtext_v2 OpenSQZ/AutoMathText-V2 (train split)

ultra_fineweb_l3 openbmb/Ultra-FineWeb-L3… See the full description on the dataset page: https://huggingface.co/datasets/vuhaian/24_collected.