explainbility_benchmark / pages /2_Explanation_Generation.py
Zekun Wu
update
394d4b8
raw
history blame
5.05 kB
import pandas as pd
import streamlit as st
from util.assistants import GPTAgent
import os
# Function to generate explanations based on a template
def generate_explanations(model_name, questions, template, temperature, max_tokens, custom_template=None):
agent = GPTAgent(model_name)
explanations = []
progress_bar = st.progress(0)
total_questions = len(questions)
for i, question in enumerate(questions):
if template == "Chain of Thought":
prompt = f"""Generate an explanation using the Chain of Thought template for the following question:
Question: {question}
Let's think step by step.
Explanation:
"""
elif template == "Custom" and custom_template:
prompt = custom_template.replace("{question}", question)
else:
prompt = f"""Generate an explanation for the following question:
Question: {question}
Explanation:
"""
response = agent.invoke(prompt, temperature=temperature, max_tokens=max_tokens).strip()
explanations.append(response)
# Update progress bar
progress_bar.progress((i + 1) / total_questions)
return explanations
# Predefined examples
examples = {
'good': {
'question': "What causes rainbows to appear in the sky?",
'explanation': "Rainbows appear when sunlight is refracted, dispersed, and reflected inside water droplets in the atmosphere, resulting in a spectrum of light appearing in the sky."
},
'bad': {
'question': "What causes rainbows to appear in the sky?",
'explanation': "Rainbows happen because light in the sky gets mixed up and sometimes shows colors when it's raining or when there is water around."
}
}
# Function to check password
def check_password():
def password_entered():
if password_input == os.getenv('PASSWORD'):
st.session_state['password_correct'] = True
else:
st.error("Incorrect Password, please try again.")
password_input = st.text_input("Enter Password:", type="password")
submit_button = st.button("Submit", on_click=password_entered)
if submit_button and not st.session_state.get('password_correct', False):
st.error("Please enter a valid password to access the demo.")
# Title of the application
st.title('Explanation Generation')
# Sidebar description of the application
st.sidebar.write("""
### Welcome to the Explanation Generation Demo
This application allows you to generate high-quality explanations for various questions using different templates. Upload a CSV of questions, select an explanation template, and generate explanations.
""")
# Check if password has been validated
if not st.session_state.get('password_correct', False):
check_password()
else:
st.sidebar.success("Password Verified. Proceed with the demo.")
st.write("""
### Instructions for Uploading CSV
Please upload a CSV file with the following column:
- `question`: The question you want explanations for.
**Example CSV Format:**
""")
# Display an example DataFrame
example_data_gen = {
"question": [
"What causes rainbows to appear in the sky?",
"Why is the sky blue?"
]
}
example_df_gen = pd.DataFrame(example_data_gen)
st.dataframe(example_df_gen)
uploaded_file_gen = st.file_uploader("Upload CSV file with 'question' column", type=['csv'])
if uploaded_file_gen is not None:
template = st.selectbox("Select an explanation template", ["Default", "Chain of Thought", "Custom"])
model_name = st.selectbox('Select a model:', ['gpt4-1106', 'gpt35-1106'])
temperature = st.sidebar.slider('Temperature', min_value=0.0, max_value=1.0, value=0.8)
max_tokens = st.sidebar.slider('Max Tokens', min_value=50, max_value=500, value=150)
custom_template = ""
if template == "Custom":
custom_template = st.text_area("Enter your custom template",
value="Generate an explanation for the following question:\n\nQuestion: {question}\n\nExplanation:")
if st.button('Generate Explanations'):
questions_df = pd.read_csv(uploaded_file_gen)
questions = questions_df['question'].tolist()
explanations = generate_explanations(model_name, questions, template, temperature, max_tokens, custom_template)
result_df_gen = pd.DataFrame({
'question': questions,
'explanation': explanations
})
st.write('### Generated Explanations')
st.dataframe(result_df_gen)
# Create a CSV download link
csv_gen = result_df_gen.to_csv(index=False)
st.download_button(
label="Download generated explanations as CSV",
data=csv_gen,
file_name='generated_explanations.csv',
mime='text/csv',
)