PocketWeights/Gliese-Qwen3.5-9B-Abliterated-WebGGUF

⚡ PocketWeights: Gliese Qwen 3.5 9B Abliterated (WebGPU & WebLLM Edition)

Heavy models, made light. PocketWeights optimizes state-of-the-art vision and language models into specialized GGUF formats designed for browser deployment, edge devices, and local consumer hardware.

🧠 About This Model

[Gliese-Qwen3.5-9B-Abliterated-Caption](https://huggingface.co/prithivMLmods/Gliese-Qwen3.5-9B-Abliterated-Caption) (by prithivMLmods) is an advanced 9-billion parameter Vision-Language Model (VLM) specialized in high-detail image captioning, visual question answering (VQA), and deep optical scene understanding.

🎯 Key Highlights of this PocketWeights Edition

• —Unrestricted Visual Understanding: Refusal directions have been orthogonally neutralized. The model produces complete, descriptive, and uncensored image captions, complex OCR extractions, and unfiltered forensic image analysis without false refusals.
• —Engineered for WebGPU & In-Browser AI (WebLLM): While modern K-quants are great for local CLI tools, WebGPU shader compilers often crash or suffer performance degradation when executing K-quant kernels. We explicitly built this collection using Q4_0, Q4_1, Q5_0, and Q5_1 for maximum cross-platform WebAssembly and browser compatibility.
• —Legacy AVX & Mobile Acceleration: Standard matrix layouts ensure fast CPU-only inference on older hardware lacking modern vector extensions.

📦 Available Files & Hardware Requirements

🚀 Beginner's Quick Start Guide

You can run this vision model locally or integrate it directly into web apps:

Option 1: Ollama (Terminal / CLI — Easiest)

Launch the model immediately in your terminal:

# Run the standard 4-bit WebGGUF
ollama run hf.co/PocketWeights/Gliese-Qwen3.5-9B-Abliterated-WebGGUF:Q4_0

# Or run the high-precision 5-bit version
ollama run hf.co/PocketWeights/Gliese-Qwen3.5-9B-Abliterated-WebGGUF:Q5_0

Option 2: LM Studio (Visual GUI)

Download and open LM Studio (Free for Windows, macOS, and Linux).

Click the Search icon on the left bar and search for:

PocketWeights/Gliese-Qwen3.5-9B-Abliterated-WebGGUF

Download Q40 or Q50, head to the Chat Tab, load the model, and attach images directly into the chat!

Option 3: WebLLM / Browser WebGPU Deployment

These weights are pre-formatted for direct integration into client-side WebGPU projects:

import * as webllm from "@mlc-ai/web-llm";

// Load Q4_0 directly into browser memory
const engine = await webllm.CreateMLCEngine(
  "PocketWeights/Gliese-Qwen3.5-9B-Abliterated-WebGGUF",
  { initProgressCallback: console.log }
);

Option 4: llama.cpp (CLI Inference)

Download your desired .gguf file and matching mmproj vision projector file from the Files and versions tab above.

Execute multimodal image analysis via CLI:

llama-minicpmv-cli -m Gliese-Qwen3.5-9B-Abliterated-WebGGUF-Q4_0.gguf --mmproj mmproj-model-f16.gguf --image test_image.jpg -p "Describe this image in extreme detail."

🤝 Support the PocketWeights Mission

I build, verify, and maintain automated pipelines to deliver lightweight, unrestricted, and hardware-friendly models to the open-source community completely for free.

Running conversion clusters, web hosting, and continuous build pipelines requires ongoing compute resources. If these models have saved you time, compute overhead, or server hosting bills, please consider supporting the project with a small tip!

☕ Donation Options

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

Web3 / Crypto (Polygon / ETH):

0x4FC189bf839A89259dd28DE8cD97883c49e15615

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

📄 License & Attribution

Base Architecture: Qwen Team / Alibaba Cloud

Finetune & Abliteration: Created by prithivMLmods

License: Apache 2.0 (Permissive open-source license for personal and commercial applications)