Submission Guidelines

Please submit your code and checkpoints as a Docker image or a .tar file of a runnable Docker container. The container should be capable of running your code and generating outputs as specified below.

– **Code Location**: The code must be located in the /workspace/submission/ folder within the Docker container. It must contain an inference script at /workspace/submission/run.py.
– **Checkpoints**: All model checkpoints or required data should be included within the Docker container and accessible by the script.

Submission Instructions

1. Build your Docker container with the necessary code and checkpoints.
2. Submit the Docker image either as:
– A Docker .tar archive file.
– A link to a pre-built Docker container image (if hosted on Docker Hub or other registries).

Submit the Docker container or image to ruibiny@alumni.cmu.edu.

Requirements

The container must contain:
– A run.py script located at /workspace/submission/run.py inside the container.
– Any necessary code or dependencies pre-installed in the container.

Input/Output Specifications

– Input:
– The script should take two input files:
– ./text_prompt.txt: Contains multiple lines of text prompts.
– ./audio_prompt.txt: Contains multiple lines of audio prompts.

– Output:
– The generated audio files should be saved in:
– ./text_prompt_output/ for audio generated from text_prompt.txt.
– ./audio_prompt_output/ for audio generated from audio_prompt.txt.

– File Naming:
– The output audio files should be named sequentially as 1.mp3, 2.mp3, etc.
– Each file should correspond to the line number of the input file, starting from line 1. For example:
– Line 1 of text_prompt.txt generates 1.mp3 in ./text_prompt_output/.
– Line 1 of audio_prompt.txt generates 1.mp3 in ./audio_prompt_output/.

Running the Container

Your Docker container should expose a method to run the script, which may look something like:

```bash docker load < your_container.tar docker run -v $(pwd)/text_prompt.txt:/workspace/submission/text_prompt.txt \ -v $(pwd)/audio_prompt.txt:/workspace/submission/audio_prompt.txt \ -v $(pwd)/text_prompt_output/:/workspace/submission/text_prompt_output/ \ -v $(pwd)/audio_prompt_output/:/workspace/submission/audio_prompt_output/ \ -w /workspace/submission \ <your_docker_image> ```

This command should:
– Mount the local text_prompt.txt and audio_prompt.txt as inputs.
– Mount the local directories ./text_prompt_output/ and ./audio_prompt_output/ to capture the generated audio.

Example Dockerfile (Optional)

Below is an example of how your Dockerfile might look. You can adjust the details according to your dependencies and framework:

```Dockerfile FROM python:3.9-slim # Set the working directory WORKDIR /workspace/submission # Copy your code and checkpoints into the container COPY . /workspace/submission # Install any necessary dependencies RUN pip install -r requirements.txt # Specify the command to run the code CMD ["python", "run.py"] ```

Notes

– Ensure the container can run fully offline with all required dependencies and model checkpoints included.
– The environment must be reproducible within the Docker container, so please ensure all dependencies are properly managed.