---
title: VideoJev-4B
canonical_url: "https://www.modelscope.ai/models/MCG-NJU/VideoJev-4B"
md_url: "https://www.modelscope.ai/models/MCG-NJU/VideoJev-4B.md"
repository: MCG-NJU/VideoJev-4B
last_updated: 2026-10-19
license: apache-2.0
model_type:
  - videojev
architectures:
  - VideoJevForVideoUnderstanding
base_model:
  - Qwen/Qwen3.5-4B
base_model_relation: finetune
parameters: 4.5B
tensor_type:
  - BF16
library_name:
  - pytorch
  - transformer
  - safetensors
frameworks:
  - pytorch
downloads: 5
stars: 0
tags:
  - video
  - video-question-answering
  - temporal-grounding
  - custom_code
---

# VideoJev-4B

> VideoJev-4B - An open-source model by MCG-NJU on ModelScope. A video System One model based on Qwen/Qwen3.5-4B, using direct candidate scoring for temporal grounding and multiple-choice video QA.

MCG-NJU/VideoJev-4B is a 4.5B-parameter machine learning model on ModelScope. licensed under apache-2.0. derived from Qwen/Qwen3.5-4B.

- **Repository**: MCG-NJU/VideoJev-4B
- **License**: apache-2.0
- **Parameters**: 4.5B
- **Base model**: Qwen/Qwen3.5-4B
- **Tags**: video, video-question-answering, temporal-grounding, custom_code
- **Downloads**: 5
- **Stars**: 0
- **Last updated**: 2026-10-19

Source: https://www.modelscope.ai/models/MCG-NJU/VideoJev-4B

---

# VideoJev-4B

A video System One model based on [Qwen/Qwen3.5-4B](https://huggingface.co/Qwen/Qwen3.5-4B), using direct candidate scoring for **temporal grounding** and **multiple-choice video QA**.

[Paper](https://arxiv.org/abs/2610.18615) · [Homepage](https://mcg-nju.github.io/VideoJev) · [Code](https://github.com/MCG-NJU/VideoJev) · [Usage guide](USAGE.md)

## Benchmark Results

![VideoJev-4B temporal grounding quality and latency](assets/benchmark-grounding.png)

![VideoJev-4B video QA quality and latency](assets/benchmark-qa.png)

## Quick start

Tested with Python 3.12, PyTorch 2.8 / CUDA 12.8, and an H200. Install a matching PyTorch build first.

```bash
pip install transformers==5.14.1 accelerate safetensors numpy pillow av==18.1.0 \
  flash-linear-attention==0.4.2 fla-core==0.4.2
pip install flash-attn causal-conv1d --no-build-isolation
```

Run grounding and QA on [Eating spaghetti](https://huggingface.co/datasets/ZhuYuhan/VideoJev-Demo/resolve/7e2b9685bca592c3b929268f6fcf82c4ad74b267/eating_spaghetti.mp4) (10 seconds), at 2 FPS.

```python
import json
import torch
from huggingface_hub import hf_hub_download
from transformers import AutoModel, AutoProcessor

video = hf_hub_download(
    repo_id="ZhuYuhan/VideoJev-Demo", filename="eating_spaghetti.mp4",
    repo_type="dataset", revision="7e2b9685bca592c3b929268f6fcf82c4ad74b267",
)
repo = "ZhuYuhan/VideoJev-4B"
processor = AutoProcessor.from_pretrained(repo, trust_remote_code=True)
model = AutoModel.from_pretrained(
    repo, trust_remote_code=True, dtype=torch.bfloat16,
    device_map={"": "cuda:0"}, attn_implementation="flash_attention_2",
)
model.prepare_for_inference()

questions = [
    {"id": "grounding-1", "task": "grounding",
     "query": "The woman in green eats pasta."},
    {"id": "qa-1", "task": "choice",
     "query": "What is the woman in green eating?",
     "options": ["A. Pasta.", "B. Rice.", "C. Soup."]},
]
results = model.predict_video(
    processor, video, questions,
    fps=2.0, max_frames=2048, context=8192,
)
print(json.dumps(results, indent=2))
```

## Output

Actual output from the video and code above:

```json
[
  {
    "id": "grounding-1",
    "task": "grounding",
    "probabilities": [
      0.005761092994362116,
      0.010865824297070503,
      0.026081746444106102,
      0.9046059250831604,
      0.9879851937294006,
      0.9887627959251404,
      0.9896195530891418,
      0.9849089980125427,
      0.833530068397522,
      0.33037981390953064
    ],
    "interval": [
      3.0,
      10.0
    ]
  },
  {
    "id": "qa-1",
    "task": "choice",
    "probabilities": [
      0.9999161958694458,
      5.755242091254331e-05,
      2.617507561808452e-05
    ],
    "answer_index": 0,
    "answer": "A. Pasta."
  }
]
```

`interval` is in seconds; `answer_index` is zero-based. `probabilities` contains independent sigmoid scores for grounding and softmax scores over QA options.

Grounding defaults to `threshold=0.25`; pass `threshold=0.4` to `predict_video` or `predict_features` to override it for one call.

See the [usage guide](USAGE.md) for task-specific settings and output details.

## Citation

```bibtex
@article{zhu2026videojev,
  title={VideoJev: System One Model for Video Understanding and Interaction},
  author={Zhu, Yuhan and Zeng, Xiangyu and Si, Qingyi and Yang, Chenxu and Li, Yifei and Qin, Luozheng and Zhou, Junhao and Yao, Dingyu and Zhang, Jun and Qin, Chuanyu and others},
  journal={arXiv preprint arXiv:2610.18615},
  year={2026}
}
```