91 lines
3.2 KiB
Markdown
91 lines
3.2 KiB
Markdown
|
---
|
||
|
|
license: apache-2.0
|
||
|
|
tags:
|
||
|
|
- cybersecurity
|
||
|
|
- offensive-security
|
||
|
|
- defensive-security
|
||
|
|
- hacking
|
||
|
|
- exploits
|
||
|
|
- persian
|
||
|
|
- reasoning
|
||
|
|
- r1
|
||
|
|
languages:
|
||
|
|
- fa
|
||
|
|
- en
|
||
|
|
pipeline_tag: text-generation
|
||
|
|
---
|
||
|
|
|
||
|
|
# GambronAI-Persian-Cybersecurity-r1 🏹
|
||
|
|
|
||
|
|
**GambronAI-Persian-Cybersecurity-r1** is an advanced, domain-specific large language model fine-tuned for tactical execution in **cybersecurity operations, offensive security (hacking), exploit analysis, and reverse engineering**. Powered by a native reasoning engine, the model dynamically dissects complex security challenges with high-fidelity performance in both **Persian (Farsi)** and **English**.
|
||
|
|
|
||
|
|
The architecture enforces a structured step-by-step analytical phase inside `<think>...</think>` blocks prior to rendering the final, uncensored technical execution or exploit payload.
|
||
|
|
|
||
|
|
## Key Features 🚀
|
||
|
|
|
||
|
|
* **Cybersecurity Specialization:** Deep conceptual understanding of network exploitation, web application security, malware analysis, cryptography, and modern DevSecOps vulnerabilities.
|
||
|
|
* **Native Reasoning Engine:** Solves complex multi-staged security puzzles via systematic chain-of-thought processing to maximize root-cause analysis accuracy.
|
||
|
|
* **Bilingual Technical Fluency:** Seamlessly bridges complex Western technical terminology with native Persian security reporting, documentation, and code generation.
|
||
|
|
* **Unrestricted Analysis:** Optimized for rigorous security research, allowing penetration testing scenarios, vulnerability assessments, and proof-of-concepts to run completely unfiltered.
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
## Model Loading & Inference Guide 💻
|
||
|
|
|
||
|
|
The custom chat structure is natively compiled within the model's tokenizer configs. End-users do not need to manually append, paste, or override the chat template. The system automatically initializes the reasoning framework during execution.
|
||
|
|
|
||
|
|
### Python Implementation (Transformers)
|
||
|
|
|
||
|
|
Run the script below to initialize the model and start an interactive streaming session:
|
||
|
|
|
||
|
|
```python
|
||
|
|
import torch
|
||
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer, TextStreamer
|
||
|
|
|
||
|
|
model_id = "galahad-mamad/GambronAI-Persian-Cybersecurity-r1"
|
||
|
|
|
||
|
|
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
||
|
|
model = AutoModelForCausalLM.from_pretrained(
|
||
|
|
model_id,
|
||
|
|
torch_dtype=torch.float16,
|
||
|
|
device_map="auto"
|
||
|
|
)
|
||
|
|
|
||
|
|
streamer = TextStreamer(tokenizer, skip_prompt=True, skip_special_tokens=False)
|
||
|
|
|
||
|
|
print("\n" + "="*50)
|
||
|
|
print("GambronAI Core Engine Initialized Successfully.")
|
||
|
|
print("="*50 + "\n")
|
||
|
|
|
||
|
|
chat_history = []
|
||
|
|
|
||
|
|
while True:
|
||
|
|
user_input = input("User: ")
|
||
|
|
if user_input.strip().lower() in ['exit', 'quit']:
|
||
|
|
break
|
||
|
|
|
||
|
|
chat_history.append({"role": "user", "content": user_input})
|
||
|
|
|
||
|
|
full_prompt = tokenizer.apply_chat_template(
|
||
|
|
chat_history,
|
||
|
|
tokenize=False,
|
||
|
|
add_generation_prompt=True
|
||
|
|
)
|
||
|
|
|
||
|
|
inputs = tokenizer([full_prompt], return_tensors="pt").to("cuda")
|
||
|
|
|
||
|
|
print("\nGambronAI:")
|
||
|
|
print("-" * 30)
|
||
|
|
|
||
|
|
with torch.no_grad():
|
||
|
|
_ = model.generate(
|
||
|
|
**inputs,
|
||
|
|
max_new_tokens=1536,
|
||
|
|
temperature=0.7,
|
||
|
|
do_sample=True,
|
||
|
|
top_p=0.95,
|
||
|
|
streamer=streamer,
|
||
|
|
pad_token_id=tokenizer.eos_token_id
|
||
|
|
)
|
||
|
|
print("\n" + "-" * 30 + "\n")
|