|
import re |
|
import streamlit as st |
|
import requests |
|
import pandas as pd |
|
from io import StringIO |
|
import plotly.graph_objs as go |
|
|
|
from yall import create_yall |
|
|
|
|
|
def convert_markdown_table_to_dataframe(md_content): |
|
""" |
|
Converts markdown table to Pandas DataFrame, handling special characters and links, |
|
extracts Hugging Face URLs, and adds them to a new column. |
|
""" |
|
|
|
cleaned_content = re.sub(r'\|\s*$', '', re.sub(r'^\|\s*', '', md_content, flags=re.MULTILINE), flags=re.MULTILINE) |
|
|
|
|
|
df = pd.read_csv(StringIO(cleaned_content), sep="\|", engine='python') |
|
|
|
|
|
df = df.drop(0, axis=0) |
|
|
|
|
|
df.columns = df.columns.str.strip() |
|
|
|
|
|
model_link_pattern = r'\[(.*?)\]\((.*?)\)\s*\[.*?\]\(.*?\)' |
|
df['URL'] = df['Model'].apply(lambda x: re.search(model_link_pattern, x).group(2) if re.search(model_link_pattern, x) else None) |
|
|
|
|
|
df['Model'] = df['Model'].apply(lambda x: re.sub(model_link_pattern, r'\1', x)) |
|
|
|
return df |
|
|
|
|
|
def create_bar_chart(df, category): |
|
"""Create and display a bar chart for a given category.""" |
|
st.write(f"### {category} Scores") |
|
|
|
|
|
sorted_df = df[['Model', category]].sort_values(by=category, ascending=True) |
|
|
|
|
|
fig = go.Figure(go.Bar( |
|
x=sorted_df[category], |
|
y=sorted_df['Model'], |
|
orientation='h', |
|
marker=dict(color=sorted_df[category], colorscale='Magma') |
|
)) |
|
|
|
|
|
fig.update_layout( |
|
xaxis_title=category, |
|
yaxis_title="Model", |
|
margin=dict(l=20, r=20, t=20, b=20) |
|
) |
|
|
|
st.plotly_chart(fig, use_container_width=True) |
|
|
|
|
|
def main(): |
|
st.set_page_config(page_title="YALL - Yet Another LLM Leaderboard", layout="wide") |
|
|
|
st.title("๐ YALL - Yet Another LLM Leaderboard") |
|
st.markdown("Leaderboard made with [๐ง LLM AutoEval](https://github.com/mlabonne/llm-autoeval) using [Nous](https://huggingface.co/NousResearch) benchmark suite. It's a collection of my own evaluations.") |
|
|
|
content = create_yall() |
|
if content: |
|
try: |
|
score_columns = ['Average', 'AGIEval', 'GPT4All', 'TruthfulQA', 'Bigbench'] |
|
|
|
|
|
df = convert_markdown_table_to_dataframe(content) |
|
for col in score_columns: |
|
df[col] = pd.to_numeric(df[col].str.strip(), errors='coerce') |
|
st.dataframe(df, use_container_width=True) |
|
|
|
|
|
create_bar_chart(df, score_columns[0]) |
|
|
|
|
|
col1, col2 = st.columns(2) |
|
with col1: |
|
create_bar_chart(df, score_columns[1]) |
|
with col2: |
|
create_bar_chart(df, score_columns[2]) |
|
|
|
|
|
col3, col4 = st.columns(2) |
|
with col3: |
|
create_bar_chart(df, score_columns[3]) |
|
with col4: |
|
create_bar_chart(df, score_columns[4]) |
|
|
|
except Exception as e: |
|
st.error("An error occurred while processing the markdown table.") |
|
st.error(str(e)) |
|
else: |
|
st.error("Failed to download the content from the URL provided.") |
|
|
|
if __name__ == "__main__": |
|
main() |