app.py

import streamlit as st
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
import torch

# Create a class for the session state
class SessionState:
    def __init__(self):
        self.conversation_history = []

# Initialize the session state
session_state = SessionState()

# Sidebar for setting parameters
st.sidebar.title("Model Parameters")
# You can add more parameters here as needed
max_length = st.sidebar.slider("Max Length", 10, 100, 50)
temperature = st.sidebar.slider("Temperature", 0.0, 1.0, 0.7)

# Load the model and tokenizer with a loading message
with st.spinner('Wait for it... the model is loading'):
    model_name = "facebook/blenderbot-400M-distill"
    model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
    tokenizer = AutoTokenizer.from_pretrained(model_name)

# Create a chat input for the user
input_text = st.chat_input("Enter your message:")

# Check if the user has entered a message
if input_text:
    # Add the user's message to the conversation history
    session_state.conversation_history.append(("User", input_text))

    # Create conversation history string
    history_string = "\n".join(message for role, message in session_state.conversation_history)

    # Tokenize the input text and history
    inputs = tokenizer.encode_plus(history_string, return_tensors="pt")
    inputs["input_ids"] = torch.cat([inputs["input_ids"], torch.tensor([tokenizer.sep_token_id])], dim=-1)
    inputs["attention_mask"] = torch.cat([inputs["attention_mask"], torch.tensor([1])], dim=-1)
    inputs = tokenizer.encode_plus(input_text, return_tensors="pt", add_special_tokens=False)
    inputs["input_ids"] = torch.cat([inputs["input_ids"], inputs["input_ids"]], dim=-1)
    inputs["attention_mask"] = torch.cat([inputs["attention_mask"], inputs["attention_mask"]], dim=-1)

    # Generate the response from the model with additional parameters
    outputs = model.generate(**inputs, max_length=max_length, do_sample=True, temperature=temperature)

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

    # Add the model's response to the conversation history
    session_state.conversation_history.append(("Assistant", response))

    # Display the conversation history using st.chat
    for role, message in session_state.conversation_history:
        if role == "User":
            st.chat_message(message, is_user=True)
        else:
            st.chat_message(message, is_user=False)