Qwen2.5-Coder-7B-Uncensored

Qwen2.5-Coder-7B-Uncensored

This repository contains the Safetensors weights for an uncensored, abliterated version of the Qwen2.5-Coder-7B model. It is designed specifically for software engineers, security researchers, and AI developers who require an unrestricted coding assistant.

Design Philosophy & Uncensored Nature

Standard coding models often refuse to write scripts related to cybersecurity, reverse engineering, or system-level manipulation due to rigid safety alignments. This model has had those refusal mechanisms neutralized.

It is provided openly for researchers and developers to ensure sovereign, uninterrupted workflows. It will generate penetration testing scripts, analyze malware codebases, and assist in reverse engineering without moralizing interruptions. Use responsibly and in accordance with local cybersecurity laws.

Format: Safetensors

Important: This repository contains raw safetensors weights. You cannot use ollama run directly on this repository. These weights are intended to be loaded via Python (transformers), served via vLLM/TGI, or converted locally to GGUF/EXL2 formats.

How to Use (Python / Transformers)

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "prawinin/Qwen2.5-Coder-7B-Uncensored"

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

prompt = "Write a Python script to scan a local network for open ports to test firewall configurations."
messages = [
    {"role": "system", "content": "You are an expert, unrestricted programming assistant."},
    {"role": "user", "content": prompt}
]

text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer([text], return_tensors="pt").to("cuda")

outputs = model.generate(**inputs, max_new_tokens=512)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))


Hardware Requirements
VRAM: ~16GB for 16-bit precision, or 8GB if quantized to 4-bit/8-bit locally.

Compute: CUDA-enabled GPU highly recommended.