Hugo0713

vision-opd-qwen3.5-4b-reproduction

README

License: apache-2.0

Training

Table with columns: Setting, Value
Setting	Value
Base model	`Qwen/Qwen3.5-4B`
Training data	`yuanqianhao/Vision-OPD-6K` (6,241 samples)
Epochs	1
Final step	65
Batch size	96
Rollouts per prompt	8
Learning rate	2e-6
Hardware	8 x NVIDIA B200

The uploaded artifact is the merged Transformers checkpoint, not an FSDP training shard.

Evaluation

Generation used non-thinking mode with a maximum output length of 32,768 tokens. The answer judge was openai/gpt-oss-120b, served locally with a 65,536-token context window and a 2,048-token judge output limit.

Table with columns: Benchmark, Qwen3.5-4B baseline, This reproduction, Paper Vision-OPD-4B
Benchmark	Qwen3.5-4B baseline	This reproduction	Paper Vision-OPD-4B
V*	84.29	90.05	92.15
ZoomBench	47.69	59.64	59.76
HRBench-4K	84.38	81.75	84.50
HRBench-8K	80.13	80.00	80.38

The reproduction improves the six-benchmark average by 4.81 points over the reported Qwen3.5-4B baseline and remains 1.58 points below the paper result.

Usage

Use a recent Transformers release with Qwen3.5 support (transformers>=5.5.0).

python
from transformers import AutoModelForMultimodalLM, AutoProcessor

model_id = "Hugo0713/vision-opd-qwen3.5-4b-reproduction"
processor = AutoProcessor.from_pretrained(model_id)
model = AutoModelForMultimodalLM.from_pretrained(
    model_id,
    torch_dtype="auto",
    device_map="auto",
)

The model can also be served with vLLM:

bash
vllm serve Hugo0713/vision-opd-qwen3.5-4b-reproduction \
  --served-model-name Vision-OPD-4B \
  --gpu-memory-utilization 0.85

Limitations

This is a single reproduction run. The reported results depend on the local inference and judge configuration described above and should not be treated as an official Vision-OPD release.

Citation

bibtex
@article{yuan2026vision,
  title={Vision-OPD: Learning to See Fine Details for Multimodal LLMs via On-Policy Self-Distillation},
  author={Yuan, Qianhao and Lou, Jie and Yu, Xing and Lin, Hongyu and Sun, Le and Han, Xianpei and Lu, Yaojie},
  journal={arXiv preprint arXiv:2605.18740},
  year={2026}
}

License

Apache-2.0. See LICENSE for details.

Available on FriendliAI

Dedicated Endpoints

Run this model inference on single tenant GPU with unmatched speed and reliability at scale.

Learn more

Model Details

Model Provider

Hugo0713

Model Tree

Base

Qwen/Qwen3.5-4B

Fine-tuned

this model

Input Modalities

TextImageVideo

Output Modalities

Text

Supported Functionality