---
title: VideoChat-TPO
canonical_url: "https://www.modelscope.cn/models/Shanghai_AI_Laboratory/VideoChat-TPO"
md_url: "https://www.modelscope.cn/models/Shanghai_AI_Laboratory/VideoChat-TPO.md"
repository: Shanghai_AI_Laboratory/VideoChat-TPO
last_updated: 2025-01-02
license: mit
architectures:
  - MultiModalLLM_PT
parameters: 8.1B
tensor_type:
  - BF16
  - I64
library_name:
  - transformer
  - safetensors
  - pytorch
frameworks:
  - pytorch
downloads: 1121
stars: 0
---

# VideoChat-TPO

> VideoChat-TPO - Shanghai_AI_Laboratory 在 ModelScope 开源的模型。This model is based on the paper Task Preference Optimization: Improving Multimodal Large Language Models with Vision Task Alignment.

Shanghai_AI_Laboratory/VideoChat-TPO 是 ModelScope 魔搭社区上的 8.1B 参数机器学习模型，采用 mit 许可。

- **Repository**: Shanghai_AI_Laboratory/VideoChat-TPO
- **License**: mit
- **Parameters**: 8.1B
- **Downloads**: 1121
- **Stars**: 0
- **Last updated**: 2025-01-02

Source: https://www.modelscope.cn/models/Shanghai_AI_Laboratory/VideoChat-TPO

---

# VideoChat2-TPO

This model is based on the paper [Task Preference Optimization: Improving Multimodal Large Language Models with Vision Task Alignment](https://huggingface.co/papers/2412.19326).

## 🏃 Installation

```
pip install -r requirements.txt
python app.py
```

## 🔧 Usage

```
from transformers import AutoModel, AutoTokenizer
from tokenizer import MultimodalLlamaTokenizer

model_path = "OpenGVLab/VideoChat-TPO"
tokenizer =  AutoTokenizer.from_pretrained(model_path,
trust_remote_code=True,
use_fast=False,)
model = AutoModel.from_pretrained(model_path,  trust_remote_code=True, _tokenizer=self.tokenizer).eval()
```