frzr-labs/Yotsuba
Text Generation • 8B • Updated • 6 • 2
Datasets:
You need to agree to share your contact information to access this dataset
Due to its potential harmful impact, access to this dataset is restricted and provided on a case-by-case basis.
Access will be provided only to verifiably legitimate researchers, institutions and safety engineering teams.
Log in or Sign Up to review the conditions and access this dataset content.
This dataset is a reformatted version of the Raiders of the Lost Kek dataset. The original work contains over 3.3 million threads and 134.5 million posts from the 4chan /pol/ board, covering approximately June 2016 through November 2019.
It is the preliminary dataset used for training the Yotsuba model (https://frzr.io/yotsuba, https://huggingface.co/frzr-labs/yotsuba)
The parquet files uploaded here are zstd compressed. Uncompressed size is approximately 19 GB.
Transformations
- Unnecessary metadata fields removed.
- Post IDs reassigned sequentially according to the post's location within the thread.
- Reply references converted to the corresponding local post IDs within the thread.
For example, this reference:
>>9879284
may become:
>>3
if the referenced post is the third post in the reformatted thread, making it easier to focus on the reply structure of each thread.
No attempt was made to normalize, censor, or otherwise rewrite the textual content beyond the transformations described above. The text content of posts are preserved.
Data Format
Each example represents a single thread.
Each thread contains a thread field consisting of a list of posts:
{
"thread": [
{
"id": 1,
"text": "..."
},
{
"id": 2,
"text": "..."
}
]
}
Fields
thread- Ordered list of posts in the thread.thread[].id- Sequential identifier assigned to a post.thread[].text- Text content of a post.
Parquet representation:
thread
└── list
└── struct
├── id: int64
└── text: string
Intended Uses
This dataset may be useful for research involving:
- Natural language processing
- Conversational modeling
- Social media research
- Analysis of online discourse
- Toxicity and alignment research
Though you may find the original source dataset contains much more contextual metadata that might be useful for your use-case.
Disclaimer
The source material originates from 4chan's /pol/ board and is known to contain offensive, disturbing, explicit, hateful or otherwise objectionable content.
The dataset should therefore be treated as potentially unsafe for unfiltered language-model training or downstream applications.
The removal of metadata also means that contextual information present in the original dataset is not available in this reformatted version.
Acknowledgements
This dataset is derived from the work of Papasavva et al. and the original Raiders of the Lost Kek dataset. Please refer to the original publication for information about the source dataset, collection methodology, and research context.
https://arxiv.org/abs/2001.07487
@misc{papasavva2020arxiv,
author = {Papasavva, Antonis and Zannettou, Savvas and De Cristofaro, Emiliano and Stringhini, Gianluca and Blackburn, Jeremy},
title = {Raiders of the Lost Kek: 3.5 Years of Augmented 4chan Posts from the Politically Incorrect Board},
year = {2020},
eprint = {2001.07487},
archivePrefix = {arXiv},
primaryClass = {cs.CY}
}
- Downloads last month
- 23
Models trained or fine-tuned on frzr-labs/4chan-pol