Bio-sensing/video_captioning_ubicomp_student_challenge

TimeChat: Docker Environment Setup (CUDA 12.1, Ubuntu 22.04, Python 3.10, PyTorch 2.1.2)

📖 Overview

This project provides a Dockerized environment to run the Python-based TimeChat application with GPU acceleration, using:

• CUDA 12.1
• Ubuntu 22.04
• Python 3.10
• PyTorch 2.1.2

Original repository:
https://github.com/RenShuhuai-Andy/TimeChat

🧱 Prerequisites

• Docker and NVIDIA Container Toolkit must be installed
• A Linux machine with a compatible NVIDIA GPU and drivers
• TimeChat and ckpt directories must be present in the current working directory

📥 Clone the Repository

Clone the required repository and initialize Git LFS:

git lfs install
git clone https://huggingface.co/Bio-sensing/video_captioning_ubicomp_student_challenge

Make sure the TimeChat directory is properly set up.

🔨 Build the Docker Image

Build the Docker image using the following command:

docker build -t cuda121_ubuntu2204_python310_torch212 -f Dockerfile_cuda121_ubuntu2204_python310_torch212 .

🚀 Run the Docker Container

Run the container with GPU access and volume mounting:

docker run -it --name "timechat" --gpus all \
    -v $(pwd)/TimeChat:/home/TimeChat \
    -v $(pwd)/ckpt:/home/ckpt/ \
    -p 12354:12354 -p 12355:12355 \
    cuda121_ubuntu2204_python310_torch212 bash

💻 Launch Jupyter Lab

Inside the container, execute the following:

cd /home/
jupyter-lab --ip 0.0.0.0 --port=12354 --allow-root --no-browser --ContentsManager.allow_hidden=True &

Then, open the following URL in your browser on the host machine:

http://localhost:12354

You can verify the environment and functionality by opening and running demo.ipynb .