Qwen3-VL-8B-Instruct-abliterated-v2.0

Qwen3-VL-8B-Instruct-abliterated-v2.0 is an abliterated (v2.0) variant of Qwen3-VL-8B-Instruct, designed for Abliterated Reasoning and Captioning. This model is fine-tuned to produce highly detailed, descriptive, and reasoning-focused outputs across a wide range of visual and multimodal contexts, including complex, sensitive, or nuanced content. It supports varied image resolutions and aspect ratios while maintaining interpretive coherence and descriptive accuracy.

Key Highlights

• Abliterated / Uncensored Captioning Fine-tuned to bypass conventional content filters while preserving factual, descriptive, and reasoning-rich outputs.

• High-Fidelity Reasoning and Descriptions Generates in-depth captions and reasoning for general, artistic, technical, abstract, and low-context images.

• Robust Across Aspect Ratios Performs consistently on wide, tall, square, panoramic, and irregular image dimensions.

• Variational Detail Control Capable of generating outputs ranging from concise summaries to intricate, multi-level descriptive reasoning.

• Foundation on Qwen3-VL-8B-Instruct Architecture Built upon Qwen3-VL-8B-Instruct’s multimodal reasoning, comprehension, and instruction-following framework.

• Multilingual Output Capability Primarily outputs in English, but adaptable to multiple languages via prompt engineering.

---

Base Model Signatures:

This model has been re-sharded and optimized for the latest Transformers version from the base model: https://huggingface.co/huihui-ai/Huihui-Qwen3-VL-8B-Instruct-abliterated.

---

Quick Start with Transformers

from transformers import Qwen3VLForConditionalGeneration, AutoProcessor
from qwen_vl_utils import process_vision_info
import torch

model = Qwen3VLForConditionalGeneration.from_pretrained(
    "prithivMLmods/Qwen3-VL-8B-Instruct-abliterated-v2",
    torch_dtype="auto",
    device_map="auto"
)

processor = AutoProcessor.from_pretrained("prithivMLmods/Qwen3-VL-8B-Instruct-abliterated-v2")

messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "image",
                "image": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-VL/assets/demo.jpeg",
            },
            {"type": "text", "text": "Provide a detailed caption and reasoning for this image."},
        ],
    }
]

text = processor.apply_chat_template(
    messages, tokenize=False, add_generation_prompt=True
)
image_inputs, video_inputs = process_vision_info(messages)

inputs = processor(
    text=[text],
    images=image_inputs,
    videos=video_inputs,
    padding=True,
    return_tensors="pt",
).to("cuda")

generated_ids = model.generate(**inputs, max_new_tokens=128)

generated_ids_trimmed = [
    out_ids[len(in_ids):] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
]

output_text = processor.batch_decode(
    generated_ids_trimmed,
    skip_special_tokens=True,
    clean_up_tokenization_spaces=False
)

print(output_text)

Intended Use

This model is suited for:

• Generating detailed, unfiltered captions and reasoning for general-purpose and artistic datasets.
• Research in content moderation, red-teaming, and generative safety analysis.
• Enabling descriptive captioning and reasoning for datasets typically excluded from mainstream models.
• Creative and exploratory applications such as storytelling, visual interpretation, and multimodal reasoning.
• Captioning and reasoning for non-standard, stylized, or abstract visual content.

Limitations

• May generate explicit, sensitive, or offensive content depending on the prompt and input image.
• Not suitable for production environments that require strict content filtering or moderation.
• Output tone, style, and reasoning depth can vary depending on phrasing and visual complexity.
• May show variability in performance on synthetic or highly abstract visuals.