datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
user_prompt_domain_classification-500000x500,000 users prompts classified into domain. Classification performed by openai/gpt-oss-120b with reasoning set to medium and temperature=0, top_p=1.
Prompts sourced and randomized from various repos including:
Roman1111111/coding-prompts
kth8/user-prompts-1M
wop/just-user-prompts
trl-lib/DeepMath-103K
ianncity/General-Distillation-Prompts-1M
ianncity/VIBE-Prompts-500000x
ianncity/science-prompts-100k
m-a-p/SuperGPQA
Total completion tokens: 70 million
prompt-safety-classification
Prompt Safety Classification Dataset
This dataset comprises prompts labeled as either safe or unsafe, curated from multiple sources to support research in prompt safety classification.
Source Datasets
nvidia/Aegis-AI-Content-Safety-Dataset-2.0
allenai/wildjailbreak-r1-v2-format-filtered
PKU-Alignment/BeaverTails (training set only)
lmsys/toxic-chat (both splits)
Data Filtering
Redacted prompts have been excluded
Prompts without labels have been… See the full description on the dataset page: https://huggingface.co/datasets/agentlans/prompt-safety-classification.prompt-injection-classification
Prompt Injection Classification (EN + TR)
A balanced, labeled set for training/evaluating prompt-injection detectors: 217 injection +
80 benign prompts (Turkish + English). text, label (0/1), label_name. Used to train
turkish-prompt-injection-detector.
from datasets import load_dataset
ds = load_dataset("fevziegeyurtsevenler/prompt-injection-classification")
By AltaySec · CC-BY-4.0