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

Qwen3-VL-2B-Thinking-abliterated

Qwen3-VL-2B-Thinking-abliterated is an abliterated (v1.0) variant of Qwen3-VL-2B-Thinking, designed for Abliterated Reasoning and Captioning. This model is optimized to generate detailed, descriptive captions and reasoning outputs across a wide range of visual and multimodal contexts—including complex, sensitive, or nuanced content—while supporting diverse aspect ratios and resolutions.

Key Highlights

• Abliterated / Uncensored Captioning – Fine-tuned to bypass conventional content filters while preserving factual, descriptive, and reasoning-rich outputs.
• High-Fidelity Descriptions – Generates comprehensive captions and reasoning for general, artistic, technical, abstract, or low-context images.
• Robust Across Aspect Ratios – Performs consistently across wide, tall, square, and irregular image dimensions.
• Variational Detail Control – Capable of producing outputs ranging from concise summaries to fine-grained, intricate descriptions and reasoning.
• Foundation on Qwen3-VL-2B-Thinking Architecture – Built upon Qwen3-VL-2B-Thinking’s strong multimodal reasoning and instruction-following capabilities.
• Multilingual Output Capability – Primarily optimized for English, with adaptability for multilingual prompts through 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-2B-Thinking-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-2B-Thinking-abliterated",
    torch_dtype="auto",
    device_map="auto"
)

processor = AutoProcessor.from_pretrained("prithivMLmods/Qwen3-VL-2B-Thinking-abliterated")

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[len(inp):] for inp, out 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, uncensored captions and reasoning for general-purpose or artistic datasets.
• Research in content moderation, red-teaming, and generative safety evaluation.
• Enabling descriptive captioning and reasoning for visual datasets typically excluded from mainstream models.
• Creative applications such as storytelling, art generation, or multimodal reasoning tasks.
• Captioning and reasoning for non-standard aspect ratios and stylized visual content.

Limitations

• May produce explicit, sensitive, or offensive descriptions depending on the image content and prompts.
• Not recommended for production systems requiring strict content moderation.
• Output style, tone, and reasoning may vary based on input phrasing.
• Accuracy can fluctuate for unfamiliar, synthetic, or highly abstract visual content.