ARC-WORLD

A platform for exploring and classifying software-based artworks

To Run the project:

• Run the frontend:

Run the following command inside the folder /arc-web:

npm run dev

or

If is in a remote server:

npm run dev -- --host 0.0.0.0 --port 5174

• Run the backend:

Move to the folder '/backend':

cd /backend

And run the following command:

python -m uvicorn api:app --reload --port 8001

Organization of the folders

• /LLM_architecture: This folder contains everything related to calling the LLM. It includes the following files:

  • config.py: Configuracion usada en inference.py(file name of LLM response, maximum number of tokens per batch, among others).
  • inference.py: Main core of the project, where it calls to the LLM. It's implemented to LLAMA-3.1-70B and LLAMA-3.1-8B, depending on the amount of GPU that you have available.
  • prompt_template.py: The prompt that is send to the LLM, inside the inference.py file we build the prompt with the information bring it in the /data.
  • result.jsonl: It's the file where we are going to append the response of the LLM.
  • new_input.jsonl: It just used for testing the model, contains the format that the LLM will read.

• /src/componenets: Compenents used in the frontend

• /src/data: Here we have different files:

  • artworks.json: Where we storage the metadata of 74 artworks. This is our library.
  • reference.json: Where we storage the reference used in this reseach that represent the base of our project. This references will be show in the References page.
  • watercolor_paint_sketch.js: An example of piece of art(code) to test the model in the Classification page.

• /src/p5: The darkflowers2Sketch.js file is the artwork that is running in the homepage.

• /src/pages: The pages of the framework:

  • HomePage.jsx: The home page.
  • LibraryPage.jsx: Where we show the list of 74 artworks with 3 filters. Each card in the list represent an artwork and have the following interactions if apply, 1) Reference: Link to the online repo where the artwork was place it the first time, 2) Live Demo: how the artwork looks like, the outcome.
  • ClassificationPage.jsx: Here we can upload a single file in .js and Run the classification. This will call the backend and send the prompt to the LLM and wait for the classification response.
  • BestPracticesPage.jsx: A list of best practices of preservation of digital arts and how to document them, taken from the article "Preservation of software-based art at Tate".
  • ReferencesPage.jsx: A list of the references used for the development of the project.