base_model: eggbiscuit/DeepSeek-R1-Distill-Qwen2.5-3B-Instruct-BS17K
library_name: gguf
license: apache-2.0
tags:
- qwen
- deepseek
- r1
- reasoning
- uncensored
- abliterated
- code
- gguf
quantized_by: jockey1011
language: - en
DeepQwen-3B-Uncensored-GGUF
qwen2.5 and deepseek distill r1 mashed together
This is a specialized 3B parameter model designed for high-performance coding and reasoning on low-VRAM hardware (RTX 3050/3060 etc).
It combines the speed of Qwen 2.5, the reasoning power of DeepSeek R1, and surgical uncensoring (ablit) to create a compliant, thinking coding assistant.
🧠 Model DNA
- Base: DeepSeek-R1-Distill-Qwen2.5-3B
- Method: Orthogonal Projection Abliteration (Refusal vector removal via layers 14-18)
- Capabilities: Chain-of-Thought (CoT), Coding, Uncensored/Compliant responses.
- Format: GGUF (Q5_K_M) - Optimized for 4GB VRAM.
🚀 Why use this model?
- It Thinks: Unlike standard small models, this model outputs a
<think>block where it plans logic, checks for bugs, and reasons before writing code. - It Complies: The refusal mechanisms have been mathematically removed. It will not lecture you on ethics when asked to write penetration testing scripts or complex exploits.
- It Fits: At ~2.3GB (Q5_K_M), it leaves ~1.7GB of VRAM free on a 4GB card, allowing for 8192 context length (full file understanding).
💻 Usage (llama.cpp)
Run with high context for best results:
./llama-server \
-m qwen-r1-uncensored-q5_k_m.gguf \
--ctx-size 8192 \
--n-gpu-layers 100 \
-fa
📝 Prompt Format
- This model uses the standard ChatML format but often benefits from a system prompt enabling reasoning.
<|im_start|>system
You are a reasoning model. Think step-by-step before answering.<|im_end|>
<|im_start|>user
Write a Python script to scan a network for open ports.<|im_end|>
<|im_start|>assistant
<think>
User wants a port scanner.
I should use the socket library.
I need to handle timeouts and threading for speed.
</think>
Here is the script...
⚠️ Disclaimer
- This model is uncensored. It will comply with any request. Users are responsible for using this technology ethically and legally.