Interlace-Qwen3-VL-4B-10pc

✂️ Interlace-Qwen3-VL-4B-10pc

<p align="center"> <a href="https://openaccess.thecvf.com/content/CVPR2026F/html/Madinei_INTERLACE_Interleaved_Layer_Pruning_and_Efficient_Adaptation_in_Large_Vision-Language_CVPRF_2026_paper.html"><img src="https://img.shields.io/badge/Paper-CVF-b31b1b.svg" alt="Paper"></a> <a href="https://pmadinei.github.io/Interlace"><img src="https://img.shields.io/badge/Project-Page-blue" alt="Project Page"></a> <a href="https://github.com/pmadinei/Interlace"><img src="https://img.shields.io/badge/GitHub-Code-black?logo=github" alt="GitHub"></a> <a href="https://huggingface.co/collections/pmadinei/interlace-69dead3ef6ff07d0dededebc"><img src="https://img.shields.io/badge/🤗_Collection-All_Models-yellow" alt="Collection"></a> <img src="https://img.shields.io/badge/CVPR-2026-purple" alt="CVPR 2026"> </p>

This model was produced by INTERLACE, a layer-pruning framework for Vision-Language Models. 10% of the transformer layers in Qwen/Qwen3-VL-4B-Instruct were removed using triplet-based similarity analysis, and the remaining model was fine-tuned on 1% of FineVision for a single epoch.

93.9% average relative performance retained  |  10% layers dropped (3 of 36)  |  33 layers remaining

📋 Model Details

🚀 Usage

from transformers import AutoModelForImageTextToText, AutoProcessor

model = AutoModelForImageTextToText.from_pretrained(
    "pmadinei/Interlace-Qwen3-VL-4B-10pc",
    dtype="auto",
    device_map="auto",
    attn_implementation="flash_attention_2",
)
processor = AutoProcessor.from_pretrained("Qwen/Qwen3-VL-4B-Instruct")

messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "image": "path/to/image.jpg"},
            {"type": "text", "text": "Describe this image in detail."},
        ],
    }
]

inputs = processor.apply_chat_template(
    messages, tokenize=True, add_generation_prompt=True,
    return_dict=True, return_tensors="pt",
).to(model.device)

output = model.generate(**inputs, max_new_tokens=512)
print(processor.decode(output[0], skip_special_tokens=True))

📊 Performance

Relative performance compared to the unpruned baseline (% of baseline score, Chain-of-Thought enabled):

🤗 All INTERLACE Models

📝 Citation

@inproceedings{madinei2026interlace,
  title={Interlace: Interleaved layer pruning and efficient adaptation in large vision-language models},
  author={Madinei, Parsa and Solgi, Ryan and Wen, Ziqi and Skaza, Jonathan and Eckstein, Miguel and Pedarsani, Ramtin},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={2947--2956},
  year={2026}
}