👁 1

Qwen3-VL-4B-Thinking-Unredacted-MAX

Qwen3-VL-4B-Thinking-Unredacted-MAX is an optimized release built on top of huihui-ai/Huihui-Qwen3-VL-4B-Thinking-abliterated. This version focuses on updated packaging, improved Transformers compatibility, and stable multimodal inference behavior, while preserving the core reasoning capabilities of the original architecture. The result is a capable 4B vision-language model designed for efficient deployment, research workflows, and multimodal experimentation.

Key Highlights

• Optimized Release Structure Streamlined repository organization for easier loading, deployment, and inference workflows.

• Modern Transformers Compatibility Updated for stable integration with recent Hugging Face Transformers versions.

• 4B Thinking Vision-Language Architecture Built on Qwen3-VL-4B-Thinking, balancing reasoning strength with efficient compute requirements.

• Stable Multimodal Inference Designed for consistent performance across image-text understanding tasks.

• Efficient Caption Generation Produces structured and detailed descriptions suitable for annotation and dataset pipelines.

• Dynamic Resolution Support Retains native support for varying image resolutions and aspect ratios.

---

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-4B-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-4B-Thinking-Unredacted-MAX",
 torch_dtype="auto",
 device_map="auto"
)

processor = AutoProcessor.from_pretrained(
 "prithivMLmods/Qwen3-VL-4B-Thinking-Unredacted-MAX"
)

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 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=256)

output_text = processor.batch_decode(
 [out[len(inp):] for inp, out in zip(inputs.input_ids, generated_ids)],
 skip_special_tokens=True,
 clean_up_tokenization_spaces=False
)

print(output_text)

Intended Use

• Multimodal research and vision-language evaluation
• Image captioning and dataset generation pipelines
• Prototyping AI systems combining text and vision
• Lightweight deployment on consumer or mid-range GPUs
• Experimental workflows in multimodal understanding

Limitations & Risks

Important Note: This model inherits constraints and behavior from its base architecture.

• Output quality depends heavily on image clarity and prompt design
• May produce incomplete or inconsistent interpretations in complex scenarios
• Requires sufficient GPU memory for stable inference
• Performance varies with decoding settings and runtime optimization

Base model

Qwen/Qwen3-VL-4B-Thinking

(23)

multi-stage trained continual "abliteration" approach for the qwen3-vl series models • 5 items • Updated 2 days ago • 5