Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,104 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import re
|
2 |
+
import streamlit as st
|
3 |
+
import requests
|
4 |
+
import pandas as pd
|
5 |
+
from io import StringIO
|
6 |
+
import plotly.graph_objs as go
|
7 |
+
|
8 |
+
from yall import create_yall
|
9 |
+
|
10 |
+
|
11 |
+
def convert_markdown_table_to_dataframe(md_content):
|
12 |
+
"""
|
13 |
+
Converts markdown table to Pandas DataFrame, handling special characters and links,
|
14 |
+
extracts Hugging Face URLs, and adds them to a new column.
|
15 |
+
"""
|
16 |
+
# Remove leading and trailing | characters
|
17 |
+
cleaned_content = re.sub(r'\|\s*$', '', re.sub(r'^\|\s*', '', md_content, flags=re.MULTILINE), flags=re.MULTILINE)
|
18 |
+
|
19 |
+
# Create DataFrame from cleaned content
|
20 |
+
df = pd.read_csv(StringIO(cleaned_content), sep="\|", engine='python')
|
21 |
+
|
22 |
+
# Remove the first row after the header
|
23 |
+
df = df.drop(0, axis=0)
|
24 |
+
|
25 |
+
# Strip whitespace from column names
|
26 |
+
df.columns = df.columns.str.strip()
|
27 |
+
|
28 |
+
# Extract Hugging Face URLs and add them to a new column
|
29 |
+
model_link_pattern = r'\[(.*?)\]\((.*?)\)\s*\[.*?\]\(.*?\)'
|
30 |
+
df['URL'] = df['Model'].apply(lambda x: re.search(model_link_pattern, x).group(2) if re.search(model_link_pattern, x) else None)
|
31 |
+
|
32 |
+
# Clean Model column to have only the model link text
|
33 |
+
df['Model'] = df['Model'].apply(lambda x: re.sub(model_link_pattern, r'\1', x))
|
34 |
+
|
35 |
+
return df
|
36 |
+
|
37 |
+
|
38 |
+
def create_bar_chart(df, category):
|
39 |
+
"""Create and display a bar chart for a given category."""
|
40 |
+
st.write(f"### {category} Scores")
|
41 |
+
|
42 |
+
# Sort the DataFrame based on the category score
|
43 |
+
sorted_df = df[['Model', category]].sort_values(by=category, ascending=True)
|
44 |
+
|
45 |
+
# Create the bar chart with color gradient
|
46 |
+
fig = go.Figure(go.Bar(
|
47 |
+
x=sorted_df[category],
|
48 |
+
y=sorted_df['Model'],
|
49 |
+
orientation='h',
|
50 |
+
marker=dict(color=sorted_df[category], colorscale='Magma')
|
51 |
+
))
|
52 |
+
|
53 |
+
# Update layout for better readability
|
54 |
+
fig.update_layout(
|
55 |
+
xaxis_title=category,
|
56 |
+
yaxis_title="Model",
|
57 |
+
margin=dict(l=20, r=20, t=20, b=20)
|
58 |
+
)
|
59 |
+
|
60 |
+
st.plotly_chart(fig, use_container_width=True)
|
61 |
+
|
62 |
+
|
63 |
+
def main():
|
64 |
+
st.set_page_config(page_title="YALL - Yet Another LLM Leaderboard", layout="wide")
|
65 |
+
|
66 |
+
st.title("π YALL - Yet Another LLM Leaderboard")
|
67 |
+
st.markdown("Leaderboard made with [π§ LLM AutoEval](https://github.com/mlabonne/llm-autoeval) using [Nous](https://huggingface.co/NousResearch) benchmark suite. It's a collection of my own evaluations.")
|
68 |
+
|
69 |
+
content = create_yall()
|
70 |
+
if content:
|
71 |
+
try:
|
72 |
+
score_columns = ['Average', 'AGIEval', 'GPT4All', 'TruthfulQA', 'Bigbench']
|
73 |
+
|
74 |
+
# Display dataframe
|
75 |
+
df = convert_markdown_table_to_dataframe(content)
|
76 |
+
for col in score_columns:
|
77 |
+
df[col] = pd.to_numeric(df[col].str.strip(), errors='coerce')
|
78 |
+
st.dataframe(df, use_container_width=True)
|
79 |
+
|
80 |
+
# Full-width plot for the first category
|
81 |
+
create_bar_chart(df, score_columns[0])
|
82 |
+
|
83 |
+
# Next two plots in two columns
|
84 |
+
col1, col2 = st.columns(2)
|
85 |
+
with col1:
|
86 |
+
create_bar_chart(df, score_columns[1])
|
87 |
+
with col2:
|
88 |
+
create_bar_chart(df, score_columns[2])
|
89 |
+
|
90 |
+
# Last two plots in two columns
|
91 |
+
col3, col4 = st.columns(2)
|
92 |
+
with col3:
|
93 |
+
create_bar_chart(df, score_columns[3])
|
94 |
+
with col4:
|
95 |
+
create_bar_chart(df, score_columns[4])
|
96 |
+
|
97 |
+
except Exception as e:
|
98 |
+
st.error("An error occurred while processing the markdown table.")
|
99 |
+
st.error(str(e))
|
100 |
+
else:
|
101 |
+
st.error("Failed to download the content from the URL provided.")
|
102 |
+
|
103 |
+
if __name__ == "__main__":
|
104 |
+
main()
|