app.py

import streamlit as st

from transformers import GPT2Tokenizer, GPT2LMHeadModel

# Function to generate a response
def generate_response(input_text):
    # Adjusted input to include the [Bot] marker
    #adjusted_input = f"{input_text} [Bot]"
    
    # Encode the adjusted input
    inputs = tokenizer(input_text, return_tensors="pt")

    # Generate a sequence of text with a slightly increased max_length to account for the prompt length
    output_sequences = model.generate(
        input_ids=inputs['input_ids'],
        attention_mask=inputs['attention_mask'],
        max_length=100,  # Adjusted max_length
        temperature=0.7,
        top_k=50,
        top_p=0.95,
        no_repeat_ngram_size=2,
        pad_token_id=tokenizer.eos_token_id,
        #early_stopping=True,
        do_sample=True
    )

    # Decode the generated sequence
    full_generated_text = tokenizer.decode(output_sequences[0], skip_special_tokens=True)

    # Extract the generated response after the [Bot] marker
    bot_response_start = full_generated_text.find('[Bot]') + len('[Bot]')
    bot_response = full_generated_text[bot_response_start:]
    
    # Trim the response to end at the last period within the specified max_length
    last_period_index = bot_response.rfind('.')
    if last_period_index != -1:
        bot_response = bot_response[:last_period_index + 1]

    return bot_response.strip()

# Load pre-trained model tokenizer (vocabulary) and model
model_name = 'KhantKyaw/Chat_GPT-2'
tokenizer = GPT2Tokenizer.from_pretrained(model_name)
model = GPT2LMHeadModel.from_pretrained(model_name)





st.title("Chat with GPT-2")

# Initialize chat history
if "messages" not in st.session_state:
    st.session_state.messages = []

# Display chat messages from history on app rerun
for message in st.session_state.messages:
    with st.container():
        st.markdown(f"**{message['role'].capitalize()}**: {message['content']}")

# React to user input
prompt = st.text_input("What is up?", key="chat_input")
if prompt:
    with st.container():
        st.markdown(f"**User**: {prompt}")
        st.session_state.messages.append({"role": "user", "content": prompt})
    


    # Decode the generated tokens and remove the eos token
    response = generate_response(prompt)

    with st.container():
        st.markdown(f"**GPT-2**: {response}")
        st.session_state.messages.append({"role": "assistant", "content": response})