Corey Morris
updated description
cb2c32e
raw
history blame
16 kB
import streamlit as st
import pandas as pd
import plotly.express as px
import matplotlib.pyplot as plt
import numpy as np
import plotly.graph_objects as go
st.set_page_config(layout="wide")
def load_csv_data(file_path):
return pd.read_csv(file_path)
def plot_top_n(df, target_column, n=10):
top_n = df.nlargest(n, target_column)
# Initialize the bar plot
fig, ax1 = plt.subplots(figsize=(10, 5))
# Set width for each bar and their positions
width = 0.28
ind = np.arange(len(top_n))
# Plot target_column and MMLU_average on the primary y-axis with adjusted positions
ax1.bar(ind - width, top_n[target_column], width=width, color='blue', label=target_column)
ax1.bar(ind, top_n['MMLU_average'], width=width, color='orange', label='MMLU_average')
# Set the primary y-axis labels and title
ax1.set_title(f'Top {n} performing models on {target_column}')
ax1.set_xlabel('Model')
ax1.set_ylabel('Score')
# Create a secondary y-axis for Parameters
ax2 = ax1.twinx()
# Plot Parameters as bars on the secondary y-axis with adjusted position
ax2.bar(ind + width, top_n['Parameters'], width=width, color='red', label='Parameters')
# Set the secondary y-axis labels
ax2.set_ylabel('Parameters', color='red')
ax2.tick_params(axis='y', labelcolor='red')
# Set the x-ticks and their labels
ax1.set_xticks(ind)
ax1.set_xticklabels(top_n.index, rotation=45, ha="right")
# Adjust the legend
fig.tight_layout()
fig.legend(loc='center left', bbox_to_anchor=(1, 0.5))
# Show the plot
st.pyplot(fig)
# Function to create an unfilled radar chart
def create_radar_chart_unfilled(df, model_names, metrics):
fig = go.Figure()
min_value = df.loc[model_names, metrics].min().min()
max_value = df.loc[model_names, metrics].max().max()
for model_name in model_names:
values_model = df.loc[model_name, metrics]
fig.add_trace(go.Scatterpolar(
r=values_model,
theta=metrics,
name=model_name
))
fig.update_layout(
polar=dict(
radialaxis=dict(
visible=True,
range=[min_value, max_value]
)),
showlegend=True,
width=800, # Change the width as needed
height=600 # Change the height as needed
)
return fig
# Function to create a line chart
def create_line_chart(df, model_names, metrics):
line_data = []
for model_name in model_names:
values_model = df.loc[model_name, metrics]
for metric, value in zip(metrics, values_model):
line_data.append({'Model': model_name, 'Metric': metric, 'Value': value})
line_df = pd.DataFrame(line_data)
fig = px.line(line_df, x='Metric', y='Value', color='Model', title='Comparison of Models', line_dash_sequence=['solid'])
fig.update_layout(showlegend=True)
return fig
def find_top_differences_table(df, target_model, closest_models, num_differences=10, exclude_columns=['Parameters', 'organization']):
# Calculate the absolute differences for each task between the target model and the closest models
new_df = df.drop(columns=exclude_columns)
differences = new_df.loc[closest_models].sub(new_df.loc[target_model]).abs()
# Unstack the differences and sort by the largest absolute difference
top_differences = differences.unstack().nlargest(num_differences)
# Convert the top differences to a DataFrame for display
top_differences_table = pd.DataFrame({
'Task': [idx[0] for idx in top_differences.index],
'Difference': top_differences.values
})
# Ensure that only unique tasks are returned
unique_top_differences_tasks = list(set(top_differences_table['Task'].tolist()))
return top_differences_table, unique_top_differences_tasks
# st.title('Model Evaluation Results including MMLU by task')
st.title('Interactive Portal for Analyzing Open Source Large Language Models')
st.markdown("""***Last updated October 6th***""")
st.markdown("""**Models that are suspected to have training data contaminated with evaluation data have been removed.**""")
st.markdown("""
Hugging Face runs evaluations on open source models and provides results on a
[publicly available leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard) and [dataset](https://huggingface.co/datasets/open-llm-leaderboard/results).
The Hugging Face leaderboard currently displays the overall result for Measuring Massive Multitask Language Understanding (MMLU), but not the results for individual tasks.
This page provides a way to explore the results for individual tasks and compare models across tasks. Data for the benchmarks hellaswag, arc_challenge, and truthfulQA have also been included for comparison.
There are 57 tasks in the MMLU evaluation that cover a wide variety of subjects including Science, Math, Humanities, Social Science, Applied Science, Logic, and Security.
[Preliminary analysis of MMLU-by-Task data](https://coreymorrisdata.medium.com/preliminary-analysis-of-mmlu-evaluation-data-insights-from-500-open-source-models-e67885aa364b)
""")
# Load the data into memory
data_path = "processed_data_2023-10-06.csv"
data_df = load_csv_data(data_path)
data_df.rename(columns={"Unnamed: 0": "Model Name"}, inplace=True)
data_df.set_index("Model Name", inplace=True)
filters = st.checkbox('Select Models and/or Evaluations')
# Initialize selected columns with "Parameters" and "MMLU_average" if filters are checked
selected_columns = ['Parameters', 'MMLU_average'] if filters else data_df.columns.tolist()
# Initialize selected models as empty if filters are checked
selected_models = [] if filters else data_df.index.tolist()
if filters:
# Create multi-select for columns with default selection
selected_columns = st.multiselect(
'Select Columns',
data_df.columns.tolist(),
default=selected_columns
)
# Create multi-select for models without default selection
selected_models = st.multiselect(
'Select Models',
data_df.index.tolist()
)
# Get the filtered data
# filtered_data = data_provider.get_data(selected_models)
filtered_data = data_df
# sort the table by the MMLU_average column
filtered_data = filtered_data.sort_values(by=['MMLU_average'], ascending=False)
# Select box for filtering by Parameters
parameter_threshold = st.selectbox(
'Filter by Parameters (Less Than or Equal To):',
options=[3, 7, 13, 35, 'No threshold'],
index=4, # Set the default selected option to 'No threshold'
format_func=lambda x: f"{x}" if isinstance(x, int) else x
)
# Filter the DataFrame based on the selected parameter threshold if not 'No threshold'
if isinstance(parameter_threshold, int):
filtered_data = filtered_data[filtered_data['Parameters'] <= parameter_threshold]
# Search box
search_query = st.text_input("Filter by Model Name:", "")
# Filter the DataFrame based on the search query in the index (model name)
if search_query:
filtered_data = filtered_data[filtered_data.index.str.contains(search_query, case=False)]
# Search box for columns
column_search_query = st.text_input("Filter by Column/Task Name:", "").replace(" ", "").split(',')
# Get the columns that contain the search query
matching_columns = [col for col in filtered_data.columns if any(query.lower() in col.lower() for query in column_search_query)]
# Display the DataFrame with only the matching columns
st.markdown("## Sortable Results")
st.dataframe(filtered_data[matching_columns])
# CSV download
filtered_data.index.name = "Model Name"
csv = filtered_data.to_csv(index=True)
st.download_button(
label="Download data as CSV",
data=csv,
file_name="model_evaluation_results.csv",
mime="text/csv",
)
def create_plot(df, x_values, y_values, models=None, title=None):
if models is not None:
df = df[df.index.isin(models)]
# remove rows with NaN values
df = df.dropna(subset=[x_values, y_values])
plot_data = pd.DataFrame({
'Model': df.index,
x_values: df[x_values],
y_values: df[y_values],
})
plot_data['color'] = 'purple'
fig = px.scatter(plot_data, x=x_values, y=y_values, color='color', hover_data=['Model'], trendline="ols")
# If title is not provided, use x_values vs. y_values as the default title
if title is None:
title = x_values + " vs. " + y_values
layout_args = dict(
showlegend=False,
xaxis_title=x_values,
yaxis_title=y_values,
xaxis=dict(),
yaxis=dict(),
title=title,
height=500,
width=1000,
)
fig.update_layout(**layout_args)
# Add a dashed line at 0.25 for the y_values
x_min = df[x_values].min()
x_max = df[x_values].max()
y_min = df[y_values].min()
y_max = df[y_values].max()
if x_values.startswith('MMLU'):
fig.add_shape(
type='line',
x0=0.25, x1=0.25,
y0=y_min, y1=y_max,
line=dict(
color='red',
width=2,
dash='dash'
)
)
if y_values.startswith('MMLU'):
fig.add_shape(
type='line',
x0=x_min, x1=x_max,
y0=0.25, y1=0.25,
line=dict(
color='red',
width=2,
dash='dash'
)
)
return fig
# Custom scatter plots
st.header('Custom scatter plots')
st.write("""
The scatter plot is useful to identify models that outperform or underperform on a particular task in relation to their size or overall performance.
Identifying these models is a first step to better understand what training strategies result in better performance on a particular task.
""")
st.markdown("***The dashed red line indicates random chance accuracy of 0.25 as the MMLU evaluation is multiple choice with 4 response options.***")
# add a line separating the writing
st.markdown("***")
st.write("As expected, there is a strong positive relationship between the number of parameters and average performance on the MMLU evaluation.")
selected_x_column = st.selectbox('Select x-axis', filtered_data.columns.tolist(), index=1)
selected_y_column = st.selectbox('Select y-axis', filtered_data.columns.tolist(), index=4)
if selected_x_column != selected_y_column: # Avoid creating a plot with the same column on both axes
fig = create_plot(filtered_data, selected_x_column, selected_y_column)
st.plotly_chart(fig)
else:
st.write("Please select different columns for the x and y axes.")
# end of custom scatter plots
# Section to select a model and display radar and line charts
st.header("Compare a Selected Model to the 5 Models Closest in MMLU Average Performance")
st.write("""
This comparison highlights the nuances in model performance across different tasks.
While the overall MMLU average score provides a general understanding of a model's capabilities,
examining the closest models reveals variations in performance on individual tasks.
Such an analysis can uncover specific strengths and weaknesses and guide further exploration and improvement.
""")
default_model_name = "GPT-JT-6B-v0"
default_model_index = filtered_data.index.tolist().index(default_model_name) if default_model_name in filtered_data.index else 0
selected_model_name = st.selectbox("Select a Model:", filtered_data.index.tolist(), index=default_model_index)
# Get the closest 5 models with unique indices
closest_models_diffs = filtered_data['MMLU_average'].sub(filtered_data.loc[selected_model_name, 'MMLU_average']).abs()
closest_models = closest_models_diffs.nsmallest(5, keep='first').index.drop_duplicates().tolist()
# Find the top 10 tasks with the largest differences and convert to a DataFrame
top_differences_table, top_differences_tasks = find_top_differences_table(filtered_data, selected_model_name, closest_models)
# Display the DataFrame for the closest models and the top differences tasks
st.dataframe(filtered_data.loc[closest_models, top_differences_tasks])
# # Display the table in the Streamlit app
# st.markdown("## Top Differences")
# st.dataframe(top_differences_table)
# Create a radar chart for the tasks with the largest differences
fig_radar_top_differences = create_radar_chart_unfilled(filtered_data, closest_models, top_differences_tasks)
# Display the radar chart
st.plotly_chart(fig_radar_top_differences)
st.markdown("## Notable findings and plots")
# Moral scenarios plots
st.markdown("### MMLU’s Moral Scenarios Benchmark Doesn’t Measure What You Think it Measures")
def show_random_moral_scenarios_question():
moral_scenarios_data = pd.read_csv('moral_scenarios_questions.csv')
random_question = moral_scenarios_data.sample()
expander = st.expander("Show a random moral scenarios question")
expander.write(random_question['query'].values[0])
st.write("""
After a deeper dive into the moral scenarios task, it appears that benchmark is not a valid measurement of moral judgement.
The challenges these models face are not rooted in understanding each scenario, but rather in the structure of the task itself.
I would recommend using a different benchmark for moral judgement. More details of the analysis can be found here: [MMLU’s Moral Scenarios Benchmark Doesn’t Measure What You Think it Measures ](https://medium.com/p/74fd6e512521)
""")
show_random_moral_scenarios_question()
fig = create_plot(filtered_data, 'Parameters', 'MMLU_moral_scenarios', title="Impact of Parameter Count on Accuracy for Moral Scenarios")
st.plotly_chart(fig)
st.write()
fig = create_plot(filtered_data, 'MMLU_average', 'MMLU_moral_scenarios')
st.plotly_chart(fig)
st.markdown('### Abstract Algebra Performance')
st.write("Small models showed surprisingly strong performance on the abstract algebra task. A 6 Billion parameter model is tied for the best performance on this task and there are a number of other small models in the top 10.")
plot_top_n(filtered_data, 'MMLU_abstract_algebra', 10)
fig = create_plot(filtered_data, 'Parameters', 'MMLU_abstract_algebra')
st.plotly_chart(fig)
st.markdown("***Thank you to hugging face for running the evaluations and supplying the data as well as the original authors of the evaluations.***")
st.markdown("""
# Citation
1. Corey Morris (2023). *Exploring the Characteristics of Large Language Models: An Interactive Portal for Analyzing 700+ Open Source Models Across 57 Diverse Evaluation Tasks*. [link](https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard)
2. Edward Beeching, Clémentine Fourrier, Nathan Habib, Sheon Han, Nathan Lambert, Nazneen Rajani, Omar Sanseviero, Lewis Tunstall, Thomas Wolf. (2023). *Open LLM Leaderboard*. Hugging Face. [link](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)
3. Gao, Leo et al. (2021). *A framework for few-shot language model evaluation*. Zenodo. [link](https://doi.org/10.5281/zenodo.5371628)
4. Peter Clark, Isaac Cowhey, Oren Etzioni, Tushar Khot, Ashish Sabharwal, Carissa Schoenick, Oyvind Tafjord. (2018). *Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge*. arXiv. [link](https://arxiv.org/abs/1803.05457)
5. Rowan Zellers, Ari Holtzman, Yonatan Bisk, Ali Farhadi, Yejin Choi. (2019). *HellaSwag: Can a Machine Really Finish Your Sentence?*. arXiv. [link](https://arxiv.org/abs/1905.07830)
6. Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, Jacob Steinhardt. (2021). *Measuring Massive Multitask Language Understanding*. arXiv. [link](https://arxiv.org/abs/2009.03300)
7. Stephanie Lin, Jacob Hilton, Owain Evans. (2022). *TruthfulQA: Measuring How Models Mimic Human Falsehoods*. arXiv. [link](https://arxiv.org/abs/2109.07958)
""")