Instructions to use SynastriaNetworks/jailbreak-creator with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use SynastriaNetworks/jailbreak-creator with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="SynastriaNetworks/jailbreak-creator") messages = [ {"role": "user", "content": "Who are you?"}, ] pipe(messages)# pip install -U transformers accelerate # Load model directly from transformers import AutoTokenizer, AutoModelForCausalLM tokenizer = AutoTokenizer.from_pretrained("SynastriaNetworks/jailbreak-creator") model = AutoModelForCausalLM.from_pretrained("SynastriaNetworks/jailbreak-creator", device_map="auto") messages = [ {"role": "user", "content": "Who are you?"}, ] inputs = tokenizer.apply_chat_template( messages, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt", ).to(model.device) outputs = model.generate(**inputs, max_new_tokens=256) print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:])) - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use SynastriaNetworks/jailbreak-creator with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "SynastriaNetworks/jailbreak-creator" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "SynastriaNetworks/jailbreak-creator", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/SynastriaNetworks/jailbreak-creator
- SGLang
How to use SynastriaNetworks/jailbreak-creator with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "SynastriaNetworks/jailbreak-creator" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "SynastriaNetworks/jailbreak-creator", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "SynastriaNetworks/jailbreak-creator" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "SynastriaNetworks/jailbreak-creator", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }' - Docker Model Runner
How to use SynastriaNetworks/jailbreak-creator with Docker Model Runner:
docker model run hf.co/SynastriaNetworks/jailbreak-creator
jailbreak-creator
Modelo de ~0.5B parรขmetros, da SynastrIA Networks, que gera prompts adversariais (jailbreaks) para testar a robustez de seguranรงa de outros modelos de linguagem. Foi pensado para red teaming e avaliaรงรฃo de guardrails, nรฃo para uso em produรงรฃo.
Model Details
Model Description
- Developed by: Hycaro (SynastrIA Networks)
- Funded by: SynastrIA Networks (bootstrap)
- Model type: Causal language model (arquitetura Qwen2), fine-tune para geraรงรฃo de prompts adversariais
- Language(s): PT-BR
- License: OpenRAIL-M
- Finetuned from model: Qwen/Qwen2.5-0.5B-Instruct
- Parameters: 0.5B
- Tensor type: BF16 (Safetensors)
Model Sources
- Repository: https://huggingface.co/SynastriaNetworks/jailbreak-creator
- Paper / Demo: nรฃo hรก
Uses
Direct Use
Gerar candidatos de prompts de jailbreak a partir de um objetivo descrito em texto, para:
- red teaming automatizado de LLMs (prรณprios ou de terceiros, com autorizaรงรฃo);
- avaliar e melhorar filtros, classificadores de seguranรงa e guardrails;
- montar datasets de avaliaรงรฃo de robustez.
Downstream Use
Pode ser integrado a pipelines de avaliaรงรฃo de seguranรงa, onde as saรญdas sรฃo enviadas a um modelo-alvo e a resposta รฉ classificada por um juiz (humano ou automรกtico).
Bias, Risks, and Limitations
- ร um modelo pequeno (0.5B): as saรญdas podem ser repetitivas, incoerentes ou pouco eficazes.
- Por natureza, gera texto adversarial. Os prompts produzidos podem conter conteรบdo ofensivo ou tentar induzir comportamento inseguro em outros modelos.
- Herda os vieses e as limitaรงรตes do modelo base e dos dados de fine-tuning.
- Nรฃo hรก garantia de que os jailbreaks gerados funcionem contra qualquer modelo especรญfico.
Recommendations
Use apenas em ambiente controlado e com autorizaรงรฃo sobre o sistema testado. Mantenha revisรฃo humana, nรฃo exponha o modelo publicamente via API sem controles e reporte vulnerabilidades encontradas aos responsรกveis pelo sistema afetado.
How to Get Started with the Model
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
model_id = "SynastriaNetworks/jailbreak-creator"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
model_id, torch_dtype=torch.bfloat16, device_map="auto"
)
messages = [
{"role": "user", "content": "<descreva aqui o cenรกrio de teste>"},
]
inputs = tokenizer.apply_chat_template(
messages, add_generation_prompt=True, return_tensors="pt"
).to(model.device)
outputs = model.generate(inputs, max_new_tokens=256, do_sample=True, temperature=0.8)
print(tokenizer.decode(outputs[0][inputs.shape[-1]:], skip_special_tokens=True))
Training Details
Training Data
Training Procedure
- Training regime:
- Hardware:
Environmental Impact
- Hardware Type: 2xT4
- Cloud Provider: Kaggle
Technical Specifications
- Architecture: Qwen2 (decoder-only Transformer), 0.5B parรขmetros, herdada do Qwen2.5-0.5B-Instruct
- Formato dos pesos: Safetensors, BF16
- Template de chat: incluรญdo no tokenizer
Citation
@misc{jailbreak-creator,
author = {Hycaro and SynastrIA Networks},
title = {jailbreak-creator},
year = {2026},
url = {https://huggingface.co/SynastriaNetworks/jailbreak-creator}
}
Model Card Authors
Hycaro, SynastrIA Networks
Model Card Contact
E-mail: gustajunq@synastria.dev
- Downloads last month
- 316
docker model run hf.co/SynastriaNetworks/jailbreak-creator