Vladimir_Putin / app.py
jeremierostan's picture
Update app.py
2d2bfa2 verified
raw
history blame
4.47 kB
import os
import gradio as gr
from anthropic import Anthropic
from pypdf import PdfReader
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.metrics.pairwise import cosine_similarity
# Set up your Anthropic API key in HF secrets
ANTHROPIC_API_KEY = os.getenv('ANTHROPIC_API_KEY')
os.environ["ANTHROPIC_API_KEY"] = ANTHROPIC_API_KEY
# Set up username and password in HF secrets
username = os.getenv('username')
password = os.getenv('password')
# Function to chunk the document
def chunk_text(text, chunk_size=1000, overlap=100):
chunks = []
start = 0
while start < len(text):
end = start + chunk_size
chunk = text[start:end]
chunks.append(chunk)
start = end - overlap
return chunks
# Function to find the most relevant chunks
def get_relevant_chunks(query, chunks, top_n=3):
vectorizer = TfidfVectorizer()
tfidf_matrix = vectorizer.fit_transform(chunks + [query])
cosine_similarities = cosine_similarity(tfidf_matrix[-1], tfidf_matrix[:-1]).flatten()
relevant_indices = cosine_similarities.argsort()[-top_n:][::-1]
return [chunks[i] for i in relevant_indices]
# Add the path to your desired knowledge base
reference_document = "Rosenshine+Principles.pdf"
reader = PdfReader(reference_document)
full_text = ''.join(page.extract_text() for page in reader.pages)
text_chunks = chunk_text(full_text)
def chat_with_assistant(message, history):
# Find relevant chunks based on the user message
relevant_chunks = get_relevant_chunks(message, text_chunks)
context = "\n".join(relevant_chunks)
# Prepare the system message
system_message = f"""You are an AI assistant answering questions based on a reference document.
You provide short, clear answers in simple language.
Use the following as context for all of your answers:
{context}
"""
# Customize instructions as needed
instructions = """
You provide useful examples and similes.
"""
system_message += instructions
# Prepare the message array
messages = []
# Add conversation history
for human_msg, ai_msg in history:
messages.append({"role": "user", "content": human_msg})
messages.append({"role": "assistant", "content": ai_msg})
# Add the current user message
messages.append({"role": "user", "content": message})
# Create Anthropic client
client = Anthropic(api_key=os.environ["ANTHROPIC_API_KEY"])
# Make the API call
response = client.messages.create(
model="claude-3-sonnet-20240307",
#model="claude-3-haiku-20240307",
max_tokens=500,
system=system_message,
messages=messages
)
return response.content[0].text.strip()
# CSS for an Anthropic-looking style
anthropic_theme = gr.themes.Default().set(
body_background_fill="#FAF9F6", # Light beige background
block_background_fill="#FFFFFF", # White for input blocks
block_title_text_color="#4A4A4A", # Dark gray for text
block_label_background_fill="#F6E3CE", # Very light orange for labels
input_background_fill="#FFFFFF", # White for input fields
button_primary_background_fill="#D97758", # Anthropic orange for primary buttons
button_primary_background_fill_hover="#8A2BE2", # Darker orange for hover
button_primary_text_color="#FFFFFF", # White text on buttons
button_secondary_background_fill="#F5D0A9", # Light orange for secondary buttons
button_secondary_background_fill_hover="#F5D0A9", # Slightly darker orange for hover
button_secondary_text_color="#4A4A4A", # Dark gray text for secondary buttons
block_border_width="1px",
block_border_color="#E0E0E0", # Light gray border
)
# Gradio interface
iface = gr.ChatInterface(
chat_with_assistant,
chatbot=gr.Chatbot(height=500),
textbox=gr.Textbox(placeholder="Type your message here...", container=False, scale=7),
# Change name and description as desired
title="Claude Custom Assistant",
description="Chat with an AI assistant powered by Claude 3.5 Sonnet, customs instructions, and a reference document",
theme=anthropic_theme,
# Change examples as desired
examples=["What are the key principles of instructional design?", "What might be barriers to learning?"],
cache_examples=True,
retry_btn=None,
undo_btn="Delete Previous",
clear_btn="Clear",
)
iface.launch(auth=(username, password))