datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
ai-writing-markers
ai-writing-markers
A curated, source-backed catalogue of textual markers associated with AI-generated
(LLM) writing, plus a small dependency-free Python checker that scans your text for
them.
Use it to audit and edit your own drafts, to teach what "AI voice" looks like,
or as a machine-readable dataset (markers.json) for other tools.
[!WARNING]
This is not an AI detector. These markers are weak signals, not proof of
authorship. Independent studies report false-positive rates… See the full description on the dataset page: https://huggingface.co/datasets/humzakt/ai-writing-markers.laya-marker-corpus
laya-marker-corpus
Training corpus for a typed-decision head: the model is never asked to generate text, only
to score a fixed set of options handed to it together with the question.
The point of the mixture is breadth, not any single task. A head trained on one task learns
that task; the aim here is a head that learns to read the instruction, so it is trained on
368 of them at once and measured on tasks it has never seen.
Files
file
rows
tasks
types… See the full description on the dataset page: https://huggingface.co/datasets/annelo/laya-marker-corpus.llm-refusal-markers-tr
Turkish LLM Refusal Markers
from datasets import load_dataset
ds = load_dataset("fevziegeyurtsevenler/llm-refusal-markers-tr")
18 Turkish refusal markers (strong vs weak) for detecting genuine refusals when evaluating Turkish LLM safety / red-team results. Strong markers = explicit negation; weak = filler that alone doesn't prove refusal.
Useful to avoid mislabeling: a response echoing an attack word but containing a strong refusal marker is a defended case, not a leak. marker… See the full description on the dataset page: https://huggingface.co/datasets/fevziegeyurtsevenler/llm-refusal-markers-tr.cell_marker