import json
import gradio as gr
import pandas as pd
from pages.summarization_playground import custom_css
css = '''
.tooltip-wrapper {
position: relative;
display: inline-block;
border-bottom: 1px dotted black;
}
.tooltip-wrapper .tooltip {
visibility: hidden;
width: 300px; /* Increased width */
background-color: black;
color: #fff;
text-align: left; /* Changed to left align */
border-radius: 6px;
padding: 5px;
position: absolute;
z-index: 1;
bottom: 125%;
left: 50%;
margin-left: -150px; /* Half of the width */
opacity: 0;
transition: opacity 0.3s;
white-space: pre-wrap; /* Allow text wrapping */
word-wrap: break-word; /* Break long words if necessary */
}
.tooltip-wrapper:hover .tooltip {
visibility: visible;
opacity: 1;
}
'''
def create_html_with_tooltip(text, tooltip):
return f'''
{text}
{tooltip}
'''
with open("prompt/prompt.json", "r") as file:
json_data = file.read()
prompts = json.loads(json_data)# Sample data for the leaderboard
winning_rate = [prompt['metric']['winning_number'] for prompt in prompts]
winning_rate = [num / sum(winning_rate) for num in winning_rate]
data = {
'Rank': [i+1 for i in range(len(prompts))],
'Methods': [create_html_with_tooltip(prompt['id'], prompt['prompt']) for prompt in prompts],
'Rouge Score': [prompt['metric']['Rouge'] for prompt in prompts],
'Winning Rate': winning_rate,
'Authors': [prompt['author'] for prompt in prompts],
'Prompts': [prompt['prompt'] for prompt in prompts]
}
df = pd.DataFrame(data)
df.sort_values(by='Rouge Score', ascending=False, inplace=True, ignore_index=True)
df['Rank'] = range(1, len(df) + 1)
# Define a list of medal emojis
medals = ['🏅', '🥈', '🥉']
for i in range(3):
df.loc[i, 'Authors'] = f"{medals[i]} {df.loc[i, 'Authors']}"
def update_leaderboard(sort_by):
sorted_df = df.sort_values(by=sort_by, ascending=False, ignore_index=True)
sorted_df['Rank'] = range(1, len(sorted_df) + 1)
# Convert DataFrame to HTML with clickable headers for sorting and without escaping
html = sorted_df.to_html(index=False, escape=False)
# Add sorting links to column headers
for column in sorted_df.columns:
html = html.replace(f'{column} | ',
f'{column} | ')
return html
def create_leaderboard():
with gr.Blocks(css=css) as demo:
gr.Markdown("# 🏆 Summarization Arena Leaderboard")
with gr.Row():
gr.Markdown("[Blog](placeholder) | [GitHub](placeholder) | [Paper](placeholder) | [Dataset](placeholder) | [Twitter](placeholder) | [Discord](placeholder)")
gr.Markdown("Welcome to our open platform for evaluating LLM summarization capabilities. We use the DATASET_NAME_PLACEHOLDER dataset to generate summaries with Qwen2-1.5b. These summaries are then evaluated by Rouge and Winning Rate from the arena")
sort_by = gr.Dropdown(list(df.columns), label="Sort by", value="Rouge Score")
gr.Markdown("**Performance**\n\n**methods**: 5, **questions**: 15")
leaderboard = gr.HTML(update_leaderboard("Rouge Score"), elem_id="leaderboard")
sort_by.change(update_leaderboard, inputs=[sort_by], outputs=[leaderboard])
return demo