• en
• zh

• prithivMLmods/Qwen3-VL-8B-Instruct-abliterated-v1
• prithivMLmods/Qwen3-VL-8B-Thinking-abliterated-v1
• prithivMLmods/Qwen3-VL-4B-Instruct-abliterated-v1
• prithivMLmods/Qwen3-VL-4B-Thinking-abliterated-v1
• prithivMLmods/Qwen3-VL-2B-Instruct-abliterated-v1
• prithivMLmods/Qwen3-VL-2B-Thinking-abliterated-v1

• text-generation-inference
• abliterated
• image-to-text
• v1.0
• llama.cpp

Qwen3-VL-abliterated-AIO-GGUF

The Qwen3-VL abliterated series from prithivMLmods comprises customized variants of Qwen3-VL 2B, 4B, and 8B Instruct and Thinking models that have undergone “abliteration” to remove or relax safety guardrails, enabling uncensored, highly detailed reasoning and captioning over a wide range of visual and multimodal content, including complex, sensitive, or nuanced images and layouts. These models are fine-tuned to produce rich, high-fidelity descriptions, step-by-step visual reasoning, and flexible output granularity (from concise summaries to intricate, multi-level narratives) while remaining robust across diverse resolutions and aspect ratios. Built on the strong multimodal understanding and instruction-following capabilities of the original Qwen3-VL Instruct and Thinking backbones, they preserve multilingual capability (primarily English, with prompt-engineered multilingual support) and are particularly suited for research, red-teaming, uncensored captioning, creative generation, and advanced visual reasoning scenarios where standard safety filters would otherwise limit outputs.

Quick Start with llama.cpp / Download using HF Download

You can run this model directly with llama.cpp or download it using the Hugging Face Hub.

Use with llama.cpp (Recommended)

Link: https://huggingface.co/prithivMLmods/Qwen3-VL-abliterated-AIO-GGUF?local-app=llama.cpp

Clone the Repository

git clone https://huggingface.co/prithivMLmods/Qwen3-VL-abliterated-AIO-GGUF

Download with hf-download

from huggingfacehub import snapshotdownload

snapshot_download(
    repo_id="prithivMLmods/Qwen3-VL-abliterated-AIO-GGUF",
    local_dir="./model",
    resume_download=True,
)

Qwen3-VL-8B-Instruct-abliterated-v1 [GGUF]

Qwen3-VL-8B-Thinking-abliterated-v1 [GGUF]

Qwen3-VL-4B-Instruct-abliterated-v1 [GGUF]

Qwen3-VL-4B-Thinking-abliterated-v1 [GGUF]

Qwen3-VL-2B-Instruct-abliterated-v1 [GGUF]

Qwen3-VL-2B-Thinking-abliterated-v1 [GGUF]

Base Model Path

prithivMLmods/Qwen3-VL-8B-Instruct-abliterated-v1

prithivMLmods/Qwen3-VL-8B-Thinking-abliterated-v1

prithivMLmods/Qwen3-VL-4B-Instruct-abliterated-v1

prithivMLmods/Qwen3-VL-4B-Thinking-abliterated-v1

prithivMLmods/Qwen3-VL-2B-Instruct-abliterated-v1

prithivMLmods/Qwen3-VL-2B-Thinking-abliterated-v1

Quants Usage

(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)

Here is a handy graph by ikawrakow comparing some lower-quality quant
types (lower is better):

!image.png