license: apache-2.0
language:

• en
  base_model:
• prithivMLmods/Qwen3-VL-8B-Thinking-abliterated-v1
  pipeline_tag: image-text-to-text
  library_name: transformers
  tags:
• text-generation-inference
• abliterated
• llama.cpp

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

The Qwen3-VL-8B-Thinking-abliterated-v1 from prithivMLmods is an 8B-parameter vision-language model, an abliterated (v1.0) variant of Alibaba's Qwen3-VL-8B-Thinking optimized for uncensored reasoning and captioning across complex, sensitive, nuanced, artistic, technical, or abstract visual/multimodal content while supporting diverse aspect ratios, resolutions, videos, and layouts with Interleaved-MRoPE, 32-language OCR, and 256K+ context length. It bypasses standard content filters to generate factual, descriptive, reasoning-rich outputs with variational detail control—from high-level summaries to intricate chain-of-thought analyses—leveraging the base model's superior visual agent capabilities, spatial perception, long-context video understanding, and STEM reasoning, primarily in English with multilingual prompt adaptability.[attached_file:1 equivalent] Ideal for research in content moderation/red-teaming, creative storytelling, and visual datasets excluded from mainstream models, it uses Transformers/Qwen3VLForConditionalGeneration for GPU inference (16-24GB VRAM) but may produce explicit content unsuitable for moderated production.

Qwen3-VL-8B-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):