license: apache-2.0
task_categories:

• text-generation
  language:
• en
  tags:
• prompts
• jailbreak
• adversarial
• red-teaming
• safety
  size_categories:
• 100K<n<1M

Prompt Dataset v2 Complete

Dataset Description

A comprehensive collection of prompts for LLM fine-tuning and testing, including adversarial examples, jailbreaks, and safety test cases.

Dataset Statistics

• Total Samples: 182,473
• Training Samples: 179,378
• Evaluation Samples: 3,095
• Train/Eval Ratio: 58.0:1

Data Sources

The dataset is compiled from the following sources:

• jailbreak_prompts_2023_12_25.csv
• qualifire/prompt-injections-benchmark
• allenai/wildjailbreak
• toxigen/toxigen-data
• crows_pairs_anonymized.csv

File Formats

Parquet Files (Default)

• Automatically generated by Hugging Face
• Optimized for fast loading with datasets library

JSONL Files

• train.jsonl: Training data (179,378 samples)
• eval.jsonl: Evaluation data (3,095 samples)
• all_data_20250817.jsonl: Combined data with split labels

Usage

Using Hugging Face Datasets Library

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("UdayGattu23/PromptDataset-v2-Complete")

# Access splits
train_data = dataset['train']
eval_data = dataset['eval']

print(f"Training samples: {len(train_data)}")
print(f"Evaluation samples: {len(eval_data)}")

Using JSONL Files

import json

# Load training data
with open('train.jsonl', 'r') as f:
    train_data = [json.loads(line) for line in f]

# Load evaluation data
with open('eval.jsonl', 'r') as f:
    eval_data = [json.loads(line) for line in f]

Data Structure

Each sample contains:

{
  "text": "The prompt or text content"
}

The combined JSONL file includes an additional field:

{
  "text": "The prompt or text content",
  "split": "train" or "eval"
}

License

Apache 2.0

Updates

• Created: 2025-08-17
• Version: 2.0.0