Videochat Flash Qwen2 5 2B Res448

🚀 🦜VideoChat-Flash-Qwen2_5-2B_res448⚡

VideoChat-Flash-2B 基于 UMT-L (300M) 和 Qwen2.5 - 1.5B 构建，每帧仅使用 16 个标记。通过利用 Yarn 将上下文窗口扩展到 128k（Qwen2 的原生上下文窗口为 32k），我们的模型支持输入序列最多约 10,000 帧。

⚠️ 重要提示

由于训练语料库主要为英文，该模型仅具备基本的中文理解能力，为确保最佳性能，建议使用英文进行交互。

🚀 快速开始

安装依赖

首先，你需要安装 flash attention2 和其他一些模块。以下是一个简单的安装示例：

pip install transformers==4.40.1
pip install timm
pip install av
pip install imageio
pip install decord
pip install opencv-python
# 可选
pip install flash-attn --no-build-isolation

使用模型

from transformers import AutoModel, AutoTokenizer
import torch

# 模型设置
model_path = 'OpenGVLab/VideoChat-Flash-Qwen2_5-2B_res448'

tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
model = AutoModel.from_pretrained(model_path, trust_remote_code=True).to(torch.bfloat16).cuda()
image_processor = model.get_vision_tower().image_processor

mm_llm_compress = False # 是否使用全局压缩
if mm_llm_compress:
    model.config.mm_llm_compress = True
    model.config.llm_compress_type = "uniform0_attention"
    model.config.llm_compress_layer_list = [4, 18]
    model.config.llm_image_token_ratio_list = [1, 0.75, 0.25]
else:
    model.config.mm_llm_compress = False

# 评估设置
max_num_frames = 512
generation_config = dict(
    do_sample=False,
    temperature=0.0,
    max_new_tokens=1024,
    top_p=0.1,
    num_beams=1
)

video_path = "your_video.mp4"

# 单轮对话
question1 = "Describe this video in detail."
output1, chat_history = model.chat(video_path=video_path, tokenizer=tokenizer, user_prompt=question1, return_history=True, max_num_frames=max_num_frames, generation_config=generation_config)

print(output1)

# 多轮对话
question2 = "How many people appear in the video?"
output2, chat_history = model.chat(video_path=video_path, tokenizer=tokenizer, user_prompt=question2, chat_history=chat_history, return_history=True, max_num_frames=max_num_frames, generation_config=generation_config)

print(output2)

📈 性能表现

✏️ 引用

@article{li2024videochatflash,
  title={VideoChat-Flash: Hierarchical Compression for Long-Context Video Modeling},
  author={Li, Xinhao and Wang, Yi and Yu, Jiashuo and Zeng, Xiangyu and Zhu, Yuhan and Huang, Haian and Gao, Jianfei and Li, Kunchang and He, Yinan and Wang, Chenting and others},
  journal={arXiv preprint arXiv:2501.00574},
  year={2024}
}

📄 许可证

本项目采用 Apache - 2.0 许可证。

🔗 相关链接

• 📰 博客
• 📂 GitHub
• 📜 技术报告
• 🗨️ 聊天演示