app.py

#""" Simple Chatbot
#@author: Nigel Gebodh
#@email: [email protected]
#"""
""" Simple Chatbot
@author: Wedyan2023
@email: [email protected]
"""

import numpy as np
import streamlit as st
from openai import OpenAI
import os
from dotenv import load_dotenv
import random
os.environ["BROWSER_GATHERUSAGESTATS"] = "false"

load_dotenv()
## Embedding Using Huggingface
#huggingface_embeddings=HuggingFaceBgeEmbeddings(
    #model_name="BAAI/bge-small-en-v1.5",      #sentence-transformers/all-MiniLM-l6-v2
    #model_kwargs={'device':'cpu'},
    #encode_kwargs={'normalize_embeddings':True}

#)

# Initialize the client
client = OpenAI(
    base_url="https://api-inference.huggingface.co/v1",
    api_key=os.environ.get('HF_TOKEN')  # Add your Huggingface token here
)

# Supported models
model_links = {
    "Meta-Llama-3-8B": "meta-llama/Meta-Llama-3-8B-Instruct"
}

# Random dog images for error messages
random_dog = [
    "0f476473-2d8b-415e-b944-483768418a95.jpg",
    "1bd75c81-f1d7-4e55-9310-a27595fa8762.jpg",
    "526590d2-8817-4ff0-8c62-fdcba5306d02.jpg",
    "1326984c-39b0-492c-a773-f120d747a7e2.jpg"
]

# Reset conversation
def reset_conversation():
    st.session_state.conversation = []
    st.session_state.messages = []
    return None

# Define the available models
models = [key for key in model_links.keys()]

# Sidebar for model selection
selected_model = st.sidebar.selectbox("Select Model", models)

# Temperature slider
temp_values = st.sidebar.slider('Select a temperature value', 0.0, 1.0, 0.5)

# Reset button
st.sidebar.button('Reset Chat', on_click=reset_conversation)

# Model description
st.sidebar.write(f"You're now chatting with **{selected_model}**")
st.sidebar.markdown("*Generated content may be inaccurate or false.*")

# Chat initialization
if "messages" not in st.session_state:
    st.session_state.messages = []

# Display chat messages
for message in st.session_state.messages:
    with st.chat_message(message["role"]):
        st.markdown(message["content"])

# Main logic to choose between data generation and data labeling
task_choice = st.selectbox("Choose Task", ["Data Generation", "Data Labeling"])

if task_choice == "Data Generation":
    classification_type = st.selectbox(
        "Choose Classification Type",
        ["Sentiment Analysis", "Binary Classification", "Multi-Class Classification"]
    )

    if classification_type == "Sentiment Analysis":
        st.write("Sentiment Analysis: Positive, Negative, Neutral")
        labels = ["Positive", "Negative", "Neutral"]
    elif classification_type == "Binary Classification":
        label_1 = st.text_input("Enter first class")
        label_2 = st.text_input("Enter second class")
        labels = [label_1, label_2]
    elif classification_type == "Multi-Class Classification":
        num_classes = st.slider("How many classes?", 3, 10, 3)
        labels = [st.text_input(f"Class {i+1}") for i in range(num_classes)]

    domain = st.selectbox("Choose Domain", ["Restaurant reviews", "E-commerce reviews", "Custom"])
    if domain == "Custom":
        domain = st.text_input("Specify custom domain")

    min_words = st.number_input("Minimum words per example", min_value=10, max_value=90, value=10)
    max_words = st.number_input("Maximum words per example", min_value=10, max_value=90, value=90)

    few_shot = st.radio("Do you want to use few-shot examples?", ["Yes", "No"])
    if few_shot == "Yes":
        num_examples = st.slider("How many few-shot examples?", 1, 5, 1)
        few_shot_examples = [
            {"content": st.text_area(f"Example {i+1}"), "label": st.selectbox(f"Label for example {i+1}", labels)}
            for i in range(num_examples)
        ]
    else:
        few_shot_examples = []

    # Ask the user how many examples they need
    num_to_generate = st.number_input("How many examples to generate?", min_value=1, max_value=100, value=10)

    # User prompt text field
    user_prompt = st.text_area("Enter your prompt to guide example generation", "")

    # System prompt generation
    system_prompt = f"You are a professional {classification_type.lower()} expert. Your role is to generate data for {domain}.\n\n"
    if few_shot_examples:
        system_prompt += "Use the following few-shot examples as a reference:\n"
        for example in few_shot_examples:
            system_prompt += f"Example: {example['content']} \n Label: {example['label']}\n"
    system_prompt += f"Generate {num_to_generate} unique examples with diverse phrasing.\n"
    system_prompt += f"Each example should have between {min_words} and {max_words} words.\n"
    system_prompt += f"Use the labels specified: {', '.join(labels)}.\n"
    if user_prompt:
        system_prompt += f"Additional instructions: {user_prompt}\n"

    st.write("System Prompt:")
    st.code(system_prompt)

    if st.button("Generate Examples"):
        # Generate examples by concatenating all inputs and sending it to the model
        with st.spinner("Generating..."):
            st.session_state.messages.append({"role": "system", "content": system_prompt})

            try:
                stream = client.chat.completions.create(
                    model=model_links[selected_model],
                    messages=[
                        {"role": m["role"], "content": m["content"]}
                        for m in st.session_state.messages
                    ],
                    temperature=temp_values,
                    stream=True,
                    max_tokens=3000,
                )
                response = st.write_stream(stream)
            except Exception as e:
                response = "Error during generation."
                random_dog_pick = 'https://random.dog/' + random_dog[np.random.randint(len(random_dog))]
                st.image(random_dog_pick)
                st.write(e)

            st.session_state.messages.append({"role": "assistant", "content": response})

else:
    # Data labeling workflow (for future implementation based on classification)
    st.write("Data Labeling functionality will go here.")