CoolFace
Modelpublic

PocketWeights/Qwen3-8B-abliterated-GGUF

sourceHugging Faceapache-2.0updated 1mo agoView on Hugging Face
0likes694downloads
Model Card

โšก PocketWeights: Qwen3 8B Abliterated (Uncensored GGUF)

Heavy models, made light. PocketWeights optimizes state-of-the-art open-source LLMs into efficient GGUF formats so you can run unrestricted, powerful AI locally on consumer hardware, gaming laptops, and edge devices.


๐Ÿง  About This Model

[huihui-ai/Qwen3-8B-abliterated](https://huggingface.co/huihui-ai/Qwen3-8B-abliterated) is a specialized modification of the Qwen3 8B architecture where refusal mechanisms have been systematically neutralized using activation steering and orthogonal projection techniques.

๐ŸŽฏ Key Enhancements in this PocketWeights Edition

  • โ€”Zero Refusals & Full Compliance: Designed to eliminate false-positive refusals, making it ideal for offline cybersecurity research, automated red-teaming, penetration testing analysis, and unrestricted creative workflows.
  • โ€”Preserved Foundational Reasoning: Unlike crude fine-tunes, the abliteration process removes refusal vectors while retaining core reasoning, logic, and coding capabilities.
  • โ€”Consumer GPU Ready: Quantized to fit comfortably within 6GB and 8GB VRAM graphics cards with near-lossless output quality.
โš ๏ธ Disclaimer: This model contains no built-in guardrails or safety filters. It is intended for authorized security research, red-teaming simulations, and local sandbox environments.

๐Ÿ“ฆ Available Files & Hardware Requirements

Quant FormatFile SizeRecommended VRAM / RAMBest For
Q4_K_M5.03 GB6 GB VRAM / 8 GB RAMThe Balanced Standard. Low memory usage, fast token generation, and minimal degradation. Great for RTX 3060, 4050, or Apple Silicon M-series.
Q6_K6.73 GB8 GB VRAM / 16 GB RAMNear-Lossless Quality. Retains maximum perplexity precision for nuanced security queries and complex logic. Ideal for RTX 3070, 4060, and desktop setups.

๐Ÿš€ Beginner's Quick Start Guide

Running this uncensored model offline on your machine requires zero complex setup:

Option 1: LM Studio (Visual GUI โ€” Easiest)

  1. 1.Download and open [LM Studio](https://lmstudio.ai/) (Available free for Windows, macOS, and Linux).
  2. 2.Click the Magnifying Glass (Search) icon on the left panel.
  3. 3.Paste: PocketWeights/Qwen3-8B-abliterated-GGUF
  4. 4.Click Download next to Q4KM or Q6_K, navigate to the Chat Tab, load the model at the top, and start your session!

Option 2: Ollama (Terminal / CLI)

If you use Ollama, you can launch the model instantly in your terminal:

bash
# Run standard 4-bit version
ollama run hf.co/PocketWeights/Qwen3-8B-abliterated-GGUF:Q4_K_M

# Or run the high-precision 6-bit version
ollama run hf.co/PocketWeights/Qwen3-8B-abliterated-GGUF:Q6_K

Option 3: Jan / Kobold.cpp / llama.cpp

Direct File Download: Head to the Files and versions tab above and download your desired .gguf file.

Load it directly into Jan.ai, Kobold.cpp, Text-Generation-WebUI, or execute via llama.cpp:

bash
llama-cli -m Qwen3-8B-abliterated-GGUF-Q4_K_M.gguf -p "Analyze the following security policy..."

๐Ÿค Support the PocketWeights Mission

I build, verify, and maintain automated quantization pipelines to bring lightweight, unrestricted, and hardware-friendly models to the developer and research community for free.

Maintaining conversion clusters, storage, and continuous testing workflows requires ongoing compute resources. If these weights have saved you time, compute overhead, or cloud hosting fees, please consider supporting the project with a small tip!

โ˜• Donation Options

Buy me a coffee on Ko-fi: ko-fi.com/iamvishalnarayan

Web3 / Crypto (Polygon / ETH):

0x4FC189bf839A89259dd28DE8cD97883c49e15615

Note: Sending via the Polygon network keeps transfer gas fees below $0.01!


๐Ÿ“„ License & Attribution

Abliteration Source: huihui-ai

Architecture Base: Created by the Qwen Team / Alibaba Cloud

License: Apache 2.0 (Permissive open-source license)