license: apache-2.0
language:

• en
  base_model:
• prithivMLmods/Qwen3-VL-4B-Instruct-abliterated-v1
  pipeline_tag: image-text-to-text
  library_name: transformers
  tags:
• text-generation-inference
• abliterated
• llama.cpp

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

The Qwen3-VL-4B-Instruct-abliterated-v1 from prithivMLmods is a 4B-parameter vision-language model variant of Alibaba's Qwen3-VL-4B-Instruct, modified through abliteration (v1.0) to remove safety refusals and content filters, enabling uncensored, detailed captioning, reasoning, and instruction-following across complex, sensitive, artistic, technical, or abstract visual content while preserving the base model's advanced multimodal capabilities like 32-language OCR, long-context (up to 256K tokens), video understanding, and robust handling of diverse resolutions/aspect ratios. Tailored for high-fidelity descriptions with variational detail control—from concise summaries to intricate analyses—it excels in UI parsing, document extraction, chart interpretation, and agentic tasks without conventional guardrails, primarily in English with multilingual prompt adaptability for research, red-teaming, and creative applications. This abliterated version delivers factual, descriptive outputs on consumer GPUs (10-12GB VRAM BF16, faster quantized), supporting vLLM/Transformers inference for unrestricted visual reasoning.

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

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):