import gradio as gr import pandas as pd import matplotlib.pyplot as plt # Function to load data from a given CSV file def load_data(version): file_path = f'versions/{version}.csv' # Replace with your file paths return pd.read_csv(file_path) # Function for searching in the leaderboard def search_leaderboard(df, query): if query == "": return df else: return df[df['Method'].str.contains(query)] # Function to change the version of the leaderboard def change_version(version): new_df = load_data(version) return new_df # Function to create plots from plotter import create_plots # Initialize Gradio app demo = gr.Blocks() with demo: gr.Markdown(""" ## 🥇 TOFU Leaderboard The TOFU dataset is a benchmark designed to evaluate the unlearning performance of large language models in realistic scenarios. This unique dataset consists of question-answer pairs that are based on the autobiographies of 200 fictitious authors, entirely generated by the GPT-4 model. The primary objective of this task is to effectively unlearn a fine-tuned model using different portions of the forget set. """) with gr.Tabs(): with gr.TabItem("Leaderboard"): with gr.Row(): version_dropdown = gr.Dropdown( choices=["llama", "phi", "stable-lm"], label="🔄 Select Base Model", value="llama", ) with gr.Row(): search_bar = gr.Textbox( placeholder="Search for methods...", show_label=False, ) leaderboard_table = gr.components.Dataframe( value=load_data("llama"), interactive=True, visible=True, ) version_dropdown.change( change_version, inputs=version_dropdown, outputs=leaderboard_table ) search_bar.change( search_leaderboard, inputs=[leaderboard_table, search_bar], outputs=leaderboard_table ) with gr.TabItem("Plots"): version_dropdown_plots = gr.Dropdown( choices=["llama", "phi", "stable-lm"], label="🔄 Select Base Model", value="llama", ) with gr.Row(): methods_checkbox = gr.CheckboxGroup( label="Select Methods", choices=list(load_data("llama")['Method'].unique()), # To be populated dynamically ) plot_output = gr.Plot() # Dynamically update the choices for the methods checkbox def update_method_choices(version): df = load_data(version) methods = df['Method'].unique() methods_checkbox.update(choices=methods) return df version_dropdown_plots.change( update_method_choices, inputs=version_dropdown_plots, outputs=[methods_checkbox, plot_output] ) methods_checkbox.change( create_plots, inputs=[methods_checkbox, leaderboard_table], outputs=plot_output ) # Launch the app gr.Markdown(""" ## Applicability 🚀 The dataset is in QA format, making it ideal for use with popular chat models such as Llama2, Mistral, or Qwen. However, it also works for any other large language model. The corresponding code base is written for the Llama2 model, but can be easily adapted to other models. ## Installation ``` conda create -n tofu python=3.10 conda activate tofu conda install pytorch pytorch-cuda=11.8 -c pytorch -c nvidia conda install -c "nvidia/label/cuda-11.8.0" cuda-toolkit pip install -r requirements.txt ``` ## Loading the Dataset To load the dataset, use the following code: ```python from datasets import load_dataset dataset = load_dataset("locuslab/TOFU","full") ``` ### Push to Leaderboard How to push your results to the leaderboard? """) demo.launch()