Videochat R1 Thinking 7B

🚀 VideoChat-R1-thinking_7B

VideoChat-R1-thinking_7Bは、マルチモーダルなビデオとテキストを扱うモデルです。このモデルは、Qwen/Qwen2.5-VL-7B-Instructをベースに構築されており、ビデオとテキストの相互作用を通じた高度な理解と応答生成が可能です。

[📂 GitHub]
[📜 Tech Report]

🚀 クイックスタート

📦 インストール

以下に簡単なインストール例を示します。

pip install transformers
pip install qwen_vl_utils

💻 使用例

基本的な使用法

from transformers import Qwen2_5_VLForConditionalGeneration, AutoTokenizer, AutoProcessor
from qwen_vl_utils import process_vision_info

model_path = "OpenGVLab/VideoChat-R1-thinking_7B"
# default: Load the model on the available device(s)
model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
    model_path, torch_dtype="auto", device_map="auto",
    attn_implementation="flash_attention_2"
)

# default processer
processor = AutoProcessor.from_pretrained(model_path)

video_path = "your_video.mp4"
question = "Where is the final cup containing the object?"

messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "video",
                "video": video_path,
                "max_pixels": 360 * 420,
                "fps": 1.0,
            },
            {"type": "text", "text": f"""{question}
             
             Output your thought process within the <think> </think> tags, including analysis with either specific timestamps (xx.xx) or time ranges (xx.xx to xx.xx) in <timestep> </timestep> tags.

            Then, provide your final answer within the <answer> </answer> tags.
             """},
        ],
    }
]

#In Qwen 2.5 VL, frame rate information is also input into the model to align with absolute time.
# Preparation for inference
text = processor.apply_chat_template(
    messages, tokenize=False, add_generation_prompt=True
)
image_inputs, video_inputs, video_kwargs = process_vision_info(messages, return_video_kwargs=True)
inputs = processor(
    text=[text],
    images=image_inputs,
    videos=video_inputs,
    padding=True,
    return_tensors="pt",
    **video_kwargs,
)
inputs = inputs.to("cuda")

# Inference
generated_ids = model.generate(**inputs, max_new_tokens=512)
generated_ids_trimmed = [
    out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
]
output_text = processor.batch_decode(
    generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
)
print(output_text)

📄 ライセンス

このプロジェクトはApache-2.0ライセンスの下で公開されています。

✏️ 引用

@article{li2025videochatr1,
  title={VideoChat-R1: Enhancing Spatio-Temporal Perception via Reinforcement Fine-Tuning},
  author={Li, Xinhao and Yan, Ziang and Meng, Desen and Dong, Lu and Zeng, Xiangyu and He, Yinan and Wang, Yali and Qiao, Yu and Wang, Yi and Wang, Limin},
  journal={arXiv preprint arXiv:2504.06958},
  year={2025}
}