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 (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
Quant Format
File Size
Recommended Hardware
Best Use Case
Q4_0
4.95 GB
6 GB VRAM / 8 GB RAM
WebLLM Standard. Lowest memory footprint and standard baseline for browser WebGPU deployments.
High Precision Standard. Excellent balance of visual fidelity and legacy CPU execution speed.
Q5_1
6.33 GB
8 GB VRAM / 16 GB RAM
Maximum Visual Fidelity. Captures subtle color grading, complex text/OCR, and dense visual layouts.
🚀 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:
bash
1# Run the standard 4-bit WebGGUF2ollama run hf.co/PocketWeights/Gliese-Qwen3.5-9B-Abliterated-WebGGUF:Q4_0
34# Or run the high-precision 5-bit version5ollama 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:
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!
