This was created as a Senior Project by Fletcher Li, Samuel Cook, Alexis Corona, Xochitl Villalvazo, and Joel Hottinger How to use the Turbo Brocoli webapp:
-
Ensure that you have all the necessary dependencies installed. The required dependencies can be found in the file: "projectEnvironmentSetup.bat"
-
Navigate to the correct folder in the repository called "frontEnd"
-
Run the following command in order to set up the locally hosted web app:
flask --app app run
This can be run with the "--debug" flag added to the end for easier code updating -
Open the local host instance at http://127.0.0.1:5000 in a web browser
The following webpage should appear:
-
Once on the webpage, the first step is to select a model.
To enter a model, first find the model url from huggingface.com and input it into the text box and click *Retrieve Model"
You are then able to select the model from the drop down list next to the text box
-
The next step is to select a dataset. To do this, follow the steps from 5 but instead input the URL for the dataset
-
If the dataset that you want to run is a subset of the main dataset, you will need to enter the name of this subset in the box below the dataset selection
-
To fine tune how the model will e trained, a number of parameters can be adjusted based on a number of different factors
The options available include:
- The number of epochs to run
- The learning rate
- The option to limit the dataset
- If set to True, this will automatically reduce the size of the database to 400 training samples, and 100 testing samples
- For quicker testing it is recommended this is set to True, for full tests keep this false
- The per-device training batch size
- The per-device eval batch size
Original Repo - https://github.com/fxl271/turbo-broccoli