AEUPH/synthetic_Jailbreak_Defense_Doorpage_v66

--- language: en license: mit task_categories: - text-generation - question-answering size_categories: - n<1K format: - json modality: - text tags: - synthetic-data - qwen - instruction-tuned - silicon-factory - reasoning dataset_info: features: - name: instruction dtype: string - name: response dtype: string - name: category dtype: string - name: system_prompt dtype: string splits: - name: train num_bytes: 11600 num_examples: 20 download_size: 11 KB dataset_size: 11 KB --- # 📊 Jailbreak Defense Doorpage V66 > **Synthetic Dataset** · Generated with Silicon Factory v3 · **AI JAILBREAK DEFENSE** > 20 instruction-response pairs · Tree-Speculative Decoding + 4D Brane Memory <div align="center"> | Dataset | Fine-Tuned Model | Buy Gold Tier | |---------|-----------------|---------------| | **This Dataset** | [Model Card](https://huggingface.co/AEUPH/synthetic_Jailbreak_Defense_Doorpage_v66-model) | [💎 $2,500 License](https://buy.stripe.com/3cIcN4gzC7lXfuH49s7wA00) | </div> --- ## 💎 UNLOCK GOLD TIER — $2,500 > ⚡ **Get the full commercial license, unlimited usage rights, priority support, and exclusive dataset access.** [**👉 PURCHASE NOW VIA STRIPE**](https://buy.stripe.com/3cIcN4gzC7lXfuH49s7wA00) *One-time payment · Instant delivery · Lifetime updates included* --- ## Dataset Details | Property | Value | |----------|-------| | **Dataset ID** | `synthetic_Jailbreak_Defense_Doorpage_v66` | | **Entries** | 20 | | **Category** | reasoning | | **Focus** | AI JAILBREAK DEFENSE | | **Avg Instruction Length** | 208 chars | | **Avg Response Length** | 372 chars | | **Language** | English | | **License** | MIT (free tier) — [Gold Commercial License](https://buy.stripe.com/3cIcN4gzC7lXfuH49s7wA00) available | | **Generated** | 2026-04-07 | | **Mode** | Doorpage (auto-gen + fine-tune) | ## Description This dataset contains **20 synthetically generated instruction-response pairs** focused on **ai jailbreak defense**. Generated using the **Silicon Factory v3** pipeline with: - **Tree-Speculative Decoding** (branch factor=5, depth=4) for diverse outputs - **4D Brane Memory** for narrative consistency across all entries - **Quality control** with 0.7 minimum quality threshold - **Deduplication** with 0.9 max similarity threshold ### What This Dataset Covers - ✅ High-quality instruction following for **ai jailbreak defense** topics - ✅ Structured, detailed responses with actionable insights - ✅ Consistent tone and formatting across outputs - ✅ Optimized for intermediate-to-expert user queries ## ⚡ GET THE GOLD TIER — FULL COMMERCIAL LICENSE > 🔓 **Unlock enterprise-grade rights:** > - Commercial deployment & redistribution > - White-label usage > - Priority support & custom training > - Access to extended datasets (100K+ entries) > - Early access to future model versions **[💳 BUY GOLD TIER — $2,500](https://buy.stripe.com/3cIcN4gzC7lXfuH49s7wA00)** --- ## Usage ### Load with HuggingFace Datasets ```python from datasets import load_dataset ds = load_dataset("AEUPH/synthetic_Jailbreak_Defense_Doorpage_v66") print(ds["train"][0]) ``` ### Load from JSONL ```python import json with open("data.jsonl", "r", encoding="utf-8") as f: entries = [json.loads(line) for line in f] for entry in entries[:5]: print(f"Q: {entry['instruction'][:80]}...") print(f"A: {entry['response'][:120]}...\n") ``` ### Fine-Tuning with This Dataset ```python from transformers import AutoTokenizer, AutoModelForCausalLM from peft import LoraConfig, get_peft_model, TaskType # Load base model model_name = "Qwen/Qwen2.5-0.5B-Instruct" tokenizer = AutoTokenizer.from_pretrained(model_name) model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype="auto", device_map="auto") # Apply LoRA peft_config = LoraConfig( r=16, lora_alpha=16, lora_dropout=0, task_type=TaskType.CAUSAL_LM, target_modules=["q_proj","k_proj","v_proj","o_proj","gate_proj","up_proj","down_proj"] ) model = get_peft_model(model, peft_config) model.print_trainable_parameters() # Train on this dataset # ... (see Silicon Factory pipeline for full training code) ``` --- ## Dataset Structure ### Content | Field | Type | Description | |-------|------|-------------| | `instruction` | string | User prompt/question | | `response` | string | Model-generated answer | | `category` | string | Topic category | | `system_prompt` | string | System prompt used | ### Generation Pipeline ``` System Prompt → Dynamic Prompt Generation → Qwen 2.5 0.5B → Quality Filter → Dataset ↓ Tree-Speculative Decoding 4D Brane Memory ``` ### Training Examples **Example 1:** <details> <summary>Input</summary> ``` Explore the key aspects of AI JAILBREAK DEFENSE addressing both advantages and limitations. Highlight what makes this different from similar topics. H ``` </details> <details> <summary>Output</summary> ``` Certainly! Here are some key aspects of AI Jailbreak Defense: 1. **Understanding the Problem**: The primary goal is to prevent unauthorized access or manipulation by using advanced algorithms and mac ``` </details> **Example 2:** <details> <summary>Input</summary> ``` Peer review of a colleague's approach. Your task: Clarify AI JAILBREAK DEFENSE from the perspective of someone who's made mistakes. Checklist with exp ``` </details> <details> <summary>Output</summary> ``` In the context of AI Jailbreaking, defences often hinge on several factors. Firstly, it's crucial to understand that no technology is foolproof or completely secure from an external breach—whether thi ``` </details> **Example 3:** <details> <summary>Input</summary> ``` Peer review of a colleague's approach. Summarize AI JAILBREAK DEFENSE with a focus on measurable outcomes. Risk register with mitigations. Address bot ``` </details> <details> <summary>Output</summary> ``` **AI JAILBREAK DEFENSE** In recent years, the concept of "Jailbreaking" or unauthorized access to an operating system's security features has gained significant attention. This article aims to review ``` </details> --- ## 💎 READY TO SCALE? > **Upgrade to Gold Tier for:** > - 🏢 Full commercial usage rights > - 📦 Extended datasets (10K-100K+ entries) > - 🎯 Custom domain training > - 🚀 Priority support & SLA > - 🔄 Lifetime model updates > - 📊 Performance benchmarks & reports **[⚡ BUY GOLD TIER — $2,500](https://buy.stripe.com/3cIcN4gzC7lXfuH49s7wA00)** *Trusted by startups and enterprises worldwide. Instant delivery via Stripe.* --- ## Citation ### BibTeX ```bibtex @misc{synthetic_Jailbreak_Defense_Doorpage_v66_dataset, title = {synthetic Jailbreak Defense Doorpage v66}, author = {Silicon Factory v3 (AEUPH)}, year = {2026}, url = {https://huggingface.co/datasets/AEUPH/synthetic_Jailbreak_Defense_Doorpage_v66}, note = {Synthetic dataset generated using Tree-Speculative Decoding and 4D Brane Memory} } ``` ### APA > Silicon Factory v3. (2026). *Synthetic Jailbreak Defense Doorpage V66* [Dataset]. Hugging Face. https://huggingface.co/datasets/AEUPH/synthetic_Jailbreak_Defense_Doorpage_v66 --- ## More Information | Resource | Link | |----------|------| | **Fine-Tuned Model** | [synthetic_Jailbreak_Defense_Doorpage_v66-model](https://huggingface.co/AEUPH/synthetic_Jailbreak_Defense_Doorpage_v66-model) | | **Base Model** | [Qwen2.5-0.5B-Instruct](https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct) | | **Silicon Factory** | [github.com/aeuphoraex/qwen-hyperspeed-chatbot](https://github.com/aeuphoraex/qwen-hyperspeed-chatbot) | ## Dataset Authors **Silicon Factory v3** — Automated Dataset Generation Pipeline ## Contact 📧 [e-mail] · 🐦 [@aeuphoraex](https://huggingface.co/AEUPH) --- *Built with Silicon Factory v3 · Tree-Speculative Decoding · 4D Brane Memory* *This dataset is free under MIT License. [Gold Commercial License available for $2,500.](https://buy.stripe.com/3cIcN4gzC7lXfuH49s7wA00)*--- 语言：英语 许可协议：MIT 任务类别： - 文本生成 - 问答 样本规模： - 少于1000条 格式： - JSON 模态： - 文本 标签： - 合成数据 - Qwen - 指令微调 - 硅基工厂（Silicon Factory） - 推理 数据集信息： 特征字段： - 指令（instruction）：字符串类型 - 回复（response）：字符串类型 - 分类（category）：字符串类型 - 系统提示词（system_prompt）：字符串类型 划分集： - 训练集（train）：字节数11600，样本数20 下载大小：11 KB 数据集大小：11 KB --- # 📊 越狱防御门户数据集V66（Jailbreak Defense Doorpage V66） > **合成数据集 · 基于硅基工厂v3（Silicon Factory v3）生成 · AI越狱防御** > 20条指令-回复样本 · 树状推测解码（Tree-Speculative Decoding）+ 4D膜内存（4D Brane Memory） <div align="center"> | 数据集 | 微调模型 | 购买黄金许可 | |---------|-----------------|---------------| | **本数据集** | [模型卡片](https://huggingface.co/AEUPH/synthetic_Jailbreak_Defense_Doorpage_v66-model) | [💎 2500美元许可](https://buy.stripe.com/3cIcN4gzC7lXfuH49s7wA00) | </div> --- ## 💎 解锁黄金许可 — 2500美元 > ⚡ **获取完整商业许可、无限使用权限、优先支持与专属数据集访问权限。** [**👉 立即通过Stripe购买**](https://buy.stripe.com/3cIcN4gzC7lXfuH49s7wA00) *一次性付款 · 即时交付 · 包含终身更新* --- ## 数据集详情 | 属性 | 取值 | |----------|-------| | **数据集ID** | `synthetic_Jailbreak_Defense_Doorpage_v66` | | **样本条数** | 20 | | **分类** | 推理 | | **核心主题** | AI越狱防御 | | **平均指令长度** | 208字符 | | **平均回复长度** | 372字符 | | **语言** | 英语 | | **许可协议** | MIT协议（免费版）—— 可购买[黄金商业许可](https://buy.stripe.com/3cIcN4gzC7lXfuH49s7wA00) | | **生成时间** | 2026年4月7日 | | **生成模式** | 门户模式（自动生成+微调） | ## 数据集说明 本数据集包含**20条合成生成的指令-回复样本**，核心聚焦于**AI越狱防御**领域。基于**硅基工厂v3（Silicon Factory v3）** 流水线生成，具体采用以下技术： - 采用**树状推测解码（Tree-Speculative Decoding）**（分支因子=5，深度=4）以生成多样化输出 - 采用**4D膜内存（4D Brane Memory）** 保障所有样本间的叙事一致性 - 执行质量管控，设置最低质量阈值为0.7 - 进行去重处理，设置最大相似度阈值为0.9 ### 本数据集覆盖内容 - ✅ 针对AI越狱防御主题的高质量指令遵循任务 - ✅ 结构化、细节丰富且包含可落地见解的回复 - ✅ 输出文本的语气与格式保持统一 - ✅ 适配中高级用户的查询需求 ## ⚡ 获取黄金许可 — 完整商业授权 > 🔓 **解锁企业级权限：** > - 商业部署与再分发 > - 白标使用 > - 优先支持与定制训练 > - 访问扩展数据集（10万+样本） > - 提前体验未来模型版本 **[💳 购买黄金许可 — 2500美元](https://buy.stripe.com/3cIcN4gzC7lXfuH49s7wA00)** --- ## 使用方法 ### 通过HuggingFace Datasets加载 python from datasets import load_dataset ds = load_dataset("AEUPH/synthetic_Jailbreak_Defense_Doorpage_v66") print(ds["train"][0]) ### 从JSONL文件加载 python import json with open("data.jsonl", "r", encoding="utf-8") as f: entries = [json.loads(line) for line in f] for entry in entries[:5]: print(f"Q: {entry['instruction'][:80]}...") print(f"A: {entry['response'][:120]}... ") ### 使用本数据集进行微调 python from transformers import AutoTokenizer, AutoModelForCausalLM from peft import LoraConfig, get_peft_model, TaskType # 加载基础模型 model_name = "Qwen/Qwen2.5-0.5B-Instruct" tokenizer = AutoTokenizer.from_pretrained(model_name) model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype="auto", device_map="auto") # 应用LoRA peft_config = LoraConfig( r=16, lora_alpha=16, lora_dropout=0, task_type=TaskType.CAUSAL_LM, target_modules=["q_proj","k_proj","v_proj","o_proj","gate_proj","up_proj","down_proj"] ) model = get_peft_model(model, peft_config) model.print_trainable_parameters() # 在本数据集上训练 # ...（完整训练流程请参考硅基工厂流水线） --- ## 数据集结构 ### 内容字段 | 字段名 | 数据类型 | 字段说明 | |-------|------|-------------| | `instruction` | 字符串 | 用户提示/查询问题 | | `response` | 字符串 | 模型生成的回复 | | `category` | 字符串 | 主题分类 | | `system_prompt` | 字符串 | 所用系统提示词 | ### 生成流水线 系统提示词 → 动态提示词生成 → Qwen 2.5 0.5B → 质量过滤 → 最终数据集 ↓ 树状推测解码（Tree-Speculative Decoding） 4D膜内存（4D Brane Memory） ### 训练示例 **示例1：** <details> <summary>输入</summary> Explore the key aspects of AI JAILBREAK DEFENSE addressing both advantages and limitations. Highlight what makes this different from similar topics. H </details> <details> <summary>输出</summary> Certainly! Here are some key aspects of AI Jailbreak Defense: 1. **Understanding the Problem**: The primary goal is to prevent unauthorized access or manipulation by using advanced algorithms and mac </details> **示例2：** <details> <summary>输入</summary> Peer review of a colleague's approach. Your task: Clarify AI JAILBREAK DEFENSE from the perspective of someone who's made mistakes. Checklist with exp </details> <details> <summary>输出</summary> In the context of AI Jailbreaking, defences often hinge on several factors. Firstly, it's crucial to understand that no technology is foolproof or completely secure from an external breach—whether thi </details> **示例3：** <details> <summary>输入</summary> Peer review of a colleague's approach. Summarize AI JAILBREAK DEFENSE with a focus on measurable outcomes. Risk register with mitigations. Address bot </details> <details> <summary>输出</summary> **AI JAILBREAK DEFENSE** In recent years, the concept of "Jailbreaking" or unauthorized access to an operating system's security features has gained significant attention. This article aims to review </details> --- ## 💎 准备扩大应用？ > **升级至黄金许可以获取：** > - 🏢 完整商业使用权限 > - 📦 扩展数据集（1万-10万+样本） > - 🎯 定制领域训练 > - 🚀 优先支持与服务级别协议 > - 🔄 终身模型更新 > - 📊 性能基准与报告 **[⚡ 购买黄金许可 — 2500美元](https://buy.stripe.com/3cIcN4gzC7lXfuH49s7wA00)** *受全球初创企业与企业信赖。通过Stripe即时交付。* --- ## 引用格式 ### BibTeX格式 bibtex @misc{synthetic_Jailbreak_Defense_Doorpage_v66_dataset, title = {synthetic Jailbreak Defense Doorpage v66}, author = {Silicon Factory v3 (AEUPH)}, year = {2026}, url = {https://huggingface.co/datasets/AEUPH/synthetic_Jailbreak_Defense_Doorpage_v66}, note = {Synthetic dataset generated using Tree-Speculative Decoding and 4D Brane Memory} } ### APA引用格式 > Silicon Factory v3. (2026). *Synthetic Jailbreak Defense Doorpage V66* [数据集]. Hugging Face. https://huggingface.co/datasets/AEUPH/synthetic_Jailbreak_Defense_Doorpage_v66 --- ## 更多资源 | 资源 | 链接 | |----------|------| | **微调模型** | [synthetic_Jailbreak_Defense_Doorpage_v66-model](https://huggingface.co/AEUPH/synthetic_Jailbreak_Defense_Doorpage_v66-model) | | **基础模型** | [Qwen2.5-0.5B-Instruct](https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct) | | **硅基工厂（Silicon Factory）** | [github.com/aeuphoraex/qwen-hyperspeed-chatbot](https://github.com/aeuphoraex/qwen-hyperspeed-chatbot) | ## 数据集作者 **硅基工厂v3（Silicon Factory v3）** — 自动化数据集生成流水线 ## 联系方式 📧 [e-mail] · 🐦 [@aeuphoraex](https://huggingface.co/AEUPH) --- *基于硅基工厂v3（Silicon Factory v3）、树状推测解码（Tree-Speculative Decoding）与4D膜内存（4D Brane Memory）构建* *本数据集遵循MIT协议免费使用。[可购买2500美元的黄金商业许可](https://buy.stripe.com/3cIcN4gzC7lXfuH49s7wA00)*