import streamlit as st
from huggingface_hub import InferenceClient
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
import os

# Replace 'your_huggingface_token' with your actual Hugging Face access token
access_token = os.getenv('token')

# Initialize the tokenizer and model with the Hugging Face access token
tokenizer = AutoTokenizer.from_pretrained("google/gemma-2b-it", use_auth_token=access_token)
model = AutoModelForCausalLM.from_pretrained(
    "google/gemma-2b-it",
    torch_dtype=torch.bfloat16,
    use_auth_token=access_token
)
model.eval()  # Set the model to evaluation mode

# Initialize the inference client (if needed for other API-based tasks)
client = InferenceClient(token=access_token)

def conversation_predict(input_text):
    """Generate a response for single-turn input using the model."""
    # Tokenize the input text
    input_ids = tokenizer(input_text, return_tensors="pt").input_ids

    # Generate a response with the model
    outputs = model.generate(input_ids, max_new_tokens=2048)

    # Decode and return the generated response
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

def respond():
    """Streamlit app for a multi-turn chat conversation."""
    st.title("Chat with Gemma")

    system_message = st.text_input("System message", value="You are a friendly Chatbot.")
    max_tokens = st.slider("Max new tokens", min_value=1, max_value=2048, value=512, step=1)
    temperature = st.slider("Temperature", min_value=0.1, max_value=4.0, value=0.7, step=0.1)
    top_p = st.slider("Top-p (nucleus sampling)", min_value=0.1, max_value=1.0, value=0.95, step=0.05)

    message = st.text_input("Your message")

    if message:
        response = conversation_predict(message)
        st.write(response)

if __name__ == "__main__":
    respond()