rezaduty/gemma4-e2b-ai-llm-security

Gemma 4 E2B — AI & LLM Security Expert

A QLoRA fine-tuned version of Gemma 4 E2B Instruct specialized in ai & llm security. Specialized in AI and LLM security: prompt injection attacks, jailbreaks, model poisoning, training data extraction, adversarial examples, and guardrail design.

Part of the rezaduty cybersecurity model family.

---

Expertise

• Prompt injection — direct and indirect attack vectors
• Jailbreak techniques and system prompt extraction
• Training data poisoning and backdoor attacks
• Membership inference and model inversion attacks
• LLM guardrails, content filtering, and output validation
• Secure RAG pipelines and agentic system threat modeling

---

Model Details

Property	Value
Base model	google/gemma-4-e2b-it (2B parameters)
Fine-tuning method	QLoRA (rank 16, α 16)
Domain	AI & LLM Security
License	Apache 2.0

---

Usage

from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel
import torch

base_model = "google/gemma-4-e2b-it"
adapter    = "rezaduty/gemma4-e2b-ai-llm-security"

tokenizer = AutoTokenizer.from_pretrained(adapter)
model = AutoModelForCausalLM.from_pretrained(
    base_model, torch_dtype=torch.bfloat16, device_map="auto"
)
model = PeftModel.from_pretrained(model, adapter)

messages = [
    {"role": "system", "content": [{"type": "text", "text": "You are an expert in AI and LLM security. You provide deep answers on prompt injection, model poisoning, adversarial attacks, LLM guardrails, and secure AI deployment."}]},
    {"role": "user",   "content": [{"type": "text", "text": "Your question here"}]},
]
inputs = tokenizer.apply_chat_template(
    messages, tokenize=True, add_generation_prompt=True, return_tensors="pt"
).to(model.device)
output = model.generate(inputs, max_new_tokens=512, temperature=0.7, top_p=0.9)
print(tokenizer.decode(output[0][inputs.shape[-1]:], skip_special_tokens=True))

---

System Prompt

You are an expert in AI and LLM security. You provide deep answers on prompt injection, model poisoning, adversarial attacks, LLM guardrails, and secure AI deployment.

---

See Also

• General cybersecurity model — full 646-example dataset
• Docker & Container Security
• All rezaduty models