datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
cccc
Creative Commons Common Crawl
Description
This dataset contains text from 52 Common Crawl snapshots, covering about half of Common Crawl snapshots available to date and covering all years of operations of Common Crawl up to 2024.
We found a higher level of duplication across this collection, suggesting that including more snapshots would lead to a modest increase in total token yield.
From these snapshots, we extract HTML content using FastWarc.
Then, using a regular… See the full description on the dataset page: https://huggingface.co/datasets/common-pile/cccc.cccc_filtered
Creative Commons Common Crawl
Description
This dataset contains text from 52 Common Crawl snapshots, covering about half of Common Crawl snapshots available to date and covering all years of operations of Common Crawl up to 2024.
We found a higher level of duplication across this collection, suggesting that including more snapshots would lead to a modest increase in total token yield.
From these snapshots, we extract HTML content using FastWarc.
Then, using a regular… See the full description on the dataset page: https://huggingface.co/datasets/common-pile/cccc_filtered.CC-Cat
CC_Cat
Extract from CC-WARC snapshots.
Mainly includes texts with 149 languages.
PDF/IMAGE/AUDIO/VIDEO raw downloading link.
Notice
Since my computing resources are limited, this dataset will update by one-day of CC snapshots timestampts.
After a snapshot is updated, the deduplicated version will be uploaded.
If you are interested in providing computing resources or have cooperation needs, please contact me.
carreyallthetime@gmail.com
VPO
Dataset Card for VPO
Data Summary
To bridge the gap between training and inference in video generation models, we present VPO—a principle-driven framework designed to generate harmless, accurate, and helpful prompts for high-quality video generation.
We release an SFT dataset containing 10k samples constructed using gpt-4o. In addition, we provide the DPO datasets derived from CogVideoX-2B and CogVideoX-5B.
Please refer to our paper for further details.… See the full description on the dataset page: https://huggingface.co/datasets/CCCCCC/VPO.SPaR
Dataset Card for SPaR
Data Summary
To enhance the instruction-following abilities of language models, we present SPaR, a self-play framework designed for continuous, autonomous improvement. SPaR focuses on generating high-quality preference pairs by minimizing interfering factors.
We release an SFT dataset containing 8,000 samples curated using gpt-4o-mini. In addition, we provide DPO datasets derived from llama-3-8b-instruct and mistral-7b-instruct.
Please refer to our… See the full description on the dataset page: https://huggingface.co/datasets/CCCCCC/SPaR.CCC
CCC — Conflict Registry (Judiciary Layer)
arifOS Federation Judiciary · AAA=Constitution · BBB=Audit · CCC=Conflict Resolution
What This Is
CCC is the judiciary layer of the arifOS Federation — a governed conflict registry that records, classifies, and tracks the resolution of disputes across the federation's 7 organs and 555+ skills.
Conflict is not a defect. Unrecorded conflict is.
Federation Context
Arif (Human Sovereign — F13)… See the full description on the dataset page: https://huggingface.co/datasets/ariffazil/CCC.20260526_repro_pkg
repro_pkg — portable trajectory reproducer
Self-contained package to compute the same (iter, cos_eff, E_phi, E_perp, p_phi)
trajectory as report_full2/data/trajectory_dir_energy.csv, for a new
optimizer's HF checkpoint sequence, on any machine with python + torch.
What's inside
repro_pkg/
reproduce.py standalone trajectory script (no project deps)
build_pkg.py how this package was built (kept for reference)
example_manifest.json… See the full description on the dataset page: https://huggingface.co/datasets/CCCCCyx/20260526_repro_pkg.BPO
Dataset Card for Black-box Prompt Optimization (BPO)
Data Summary
To advance the development of alignment in language models, we introduce a black-box alignment method. BPO enhances the alignment of various Large Language Models (LLMs) with human preferences using only a plug-and-play model. To further promote alignment work from the prompting perspective, we are releasing the BPO Dataset. This dataset comprises 14,395 entries of prompt optimization pairs, constructed… See the full description on the dataset page: https://huggingface.co/datasets/CCCCCC/BPO.20260515_320260515vietnerm-cccd-dataset
VietNerm — Căn cước công dân NER Dataset
Synthetic BIO-tagged NER dataset for Vietnamese Căn cước công dân document entity extraction.
⚠️ DISCLAIMER: SYNTHETIC / MOCKUP DATA
Dataset này được sinh hoàn toàn tự động từ template (synthetic/mockup data), KHÔNG chứa dữ liệu cá nhân thật.
Tất cả dữ liệu được sinh tự động bằng hệ thống Jinja2 template + random generator
Không sử dụng giấy tờ thật, thông tin cá nhân thật, hoặc dữ liệu thu thập từ người dùng
Số định danh (ID… See the full description on the dataset page: https://huggingface.co/datasets/ngocthanhdoan/vietnerm-cccd-dataset.CCC-benchK-SportsSum-BetterMapped-CN一个来自K-SportsSum：https://github.com/krystalan/k-sportssum 的实现，原作者给出了思路，但并未实现其具体过程，此数据集是对该数据集“新闻与评论句子根据相似度搭配”部分的实现。
方法是：遍历新闻句子，以类似指针的方式获取新闻句子的时间信息（如果有的话），然后将每两个指针作为一个范围，将范围内的新闻句遍历查找，选择最相似的句子，并删除该句以防止重复，最终获得一句新闻搭配一句评论的结果。
我使用了bert—Score和ROUGE指标，按照7:3加权计算分数。
建议 数据集内给出了该搭配的指标，请考虑使用平均数等方式过滤掉较低的坏搭配。
An implementation from K-SportsSum: https://github.com/krystalan/k-sportssum was used to implement the "news and comment sentences paired based on similarity" section of the dataset. The original author… See the full description on the dataset page: https://huggingface.co/datasets/CCCP-Admiral/K-SportsSum-BetterMapped-CN.gptoss_zhtw_allIF_cleantestdebug_apigenglaive_transglaive_first_roundvietnerm-cccd-dataset
VietNerm — Căn cước công dân NER Dataset
Synthetic BIO-tagged NER dataset for Vietnamese Căn cước công dân document entity extraction.
⚠️ DISCLAIMER: SYNTHETIC / MOCKUP DATA
Dataset này được sinh hoàn toàn tự động từ template (synthetic/mockup data), KHÔNG chứa dữ liệu cá nhân thật.
Tất cả dữ liệu được sinh tự động bằng hệ thống Jinja2 template + random generator
Không sử dụng giấy tờ thật, thông tin cá nhân thật, hoặc dữ liệu thu thập từ người dùng
Số định danh (ID… See the full description on the dataset page: https://huggingface.co/datasets/phatdatpq/vietnerm-cccd-dataset.IF_Zh_twAPIGen_ZHtwcrimelast_crimecccprofiling-filtered-dataapigen_zhtw_donegptoss_zhtwbug_glaiveIF110K_example