app.py

import gradio as gr
from transformers import AutoTokenizer, AutoModelForCausalLM

# Load the model and tokenizer
model, tokenizer = (
    AutoModelForCausalLM.from_pretrained("Abhishekcr448/Tiny-Hinglish-Chat-21M"),
    AutoTokenizer.from_pretrained("Abhishekcr448/Tiny-Hinglish-Chat-21M"),
)

# Function to generate text (suggestions)
def generate_text(prompt, output_length, temperature, top_k, top_p):
    inputs = tokenizer(prompt, return_tensors='pt').to(model.device)
    generated_output = model.generate(
        inputs['input_ids'],
        max_length=inputs['input_ids'].shape[-1] + output_length,  # Generate 10 more tokens
        no_repeat_ngram_size=2,
        temperature=temperature,
        top_k=top_k,
        top_p=top_p,
        do_sample=True,
    )
    output_text = tokenizer.decode(generated_output[0], skip_special_tokens=True)
    return output_text

# Set up the Gradio interface with custom CSS
with gr.Blocks(css="""
    #response-text {
        background-color: #e1bee7; /* Light purple background */
        border-radius: 8px; /* Rounded corners */
        padding: 10px; /* Padding inside the textbox */
        font-size: 16px; /* Font size */
        color: #4a148c; /* Dark purple text color */
    }
""") as demo:

    # Add a title to the interface
    gr.Markdown("# Hinglish Chat Prediction")

    # Add a chat interface above the text boxes with reduced size
    with gr.Row():
        chatbox = gr.Chatbot(label="Chat", type="messages", height=350, value=[{"role": "assistant", "content": "Kya kar rahe ho"}])
    
    with gr.Row():
        # Create a column for the two text boxes
        with gr.Column(scale=3):
            # Input text box for user input (first column)
            input_text = gr.Textbox(label="Start chatting", interactive=True)
            
        # Create a separate column for the buttons
        with gr.Column(scale=1):
            # Submit button placed above the replace button
            submit_button = gr.Button("Submit", variant="primary", elem_id="submit-btn")

    with gr.Row():
        # Create a column for the two text boxes
        with gr.Column(scale=3):
            # Response text box (second column)
            response_text = gr.Textbox(label="Suggestion", interactive=False, elem_id="response-text")

        # Create a separate column for the buttons
        with gr.Column(scale=1):
            replace_button = gr.Button("Use Suggestion", variant="secondary", elem_id="replace-btn")
            regenerate_button = gr.Button("Regenerate", variant="secondary", elem_id="regenerate-btn")

    with gr.Row():
        # Create a dropdown menu for text generation parameters
        with gr.Column(scale=1):
            with gr.Accordion("Change Parameters", open=False):
                output_length_slider = gr.Slider(1, 20, value=8, label="Output Length", step=1)
                temperature_slider = gr.Slider(0.1, 1.0, value=0.8, label="Temperature (Controls randomness)")
                top_k_slider = gr.Slider(1, 100, value=50, label="Top-k (Limits vocabulary size)", step=1)
                top_p_slider = gr.Slider(0.1, 1.0, value=0.9, label="Top-p (Nucleus sampling)")
    
    # Set up the interaction between input and output
    def validate_and_generate(prompt, output_length, temperature, top_k, top_p):
        if prompt.strip():
            print(f"Prompt: {prompt}")
            return generate_text(prompt, output_length, temperature, top_k, top_p)

    input_text.input(validate_and_generate, inputs=[input_text, output_length_slider, temperature_slider, top_k_slider, top_p_slider], outputs=response_text)
    replace_button.click(lambda x: x, inputs=response_text, outputs=input_text)

    def chat_interaction(prompt, history, output_length, temperature, top_k, top_p):
        if prompt.strip():
            response = generate_text(prompt, output_length, temperature, top_k, top_p)

            # Exclude the input prompt text from the response
            response = response[len(prompt):].strip()
            history.append({"role": "user", "content": prompt})
            history.append({"role": "assistant", "content": response})
            
            # Call validate_and_generate with the response
            response_text_value = validate_and_generate(response, output_length, temperature, top_k, top_p)
            return history, response_text_value[len(response):].strip(), ""
        
        return history, "", ""
        
    def regenerate_text(input_text, history, output_length, temperature, top_k, top_p):
        if input_text.strip():
            return generate_text(input_text, output_length, temperature, top_k, top_p)
        else:
            last_message = history[-1]["content"]
            return generate_text(last_message, output_length, temperature, top_k, top_p)[len(last_message):].strip()

    input_text.submit(chat_interaction, inputs=[input_text, chatbox, output_length_slider, temperature_slider, top_k_slider, top_p_slider], outputs=[chatbox, response_text, input_text])
    submit_button.click(chat_interaction, inputs=[input_text, chatbox, output_length_slider, temperature_slider, top_k_slider, top_p_slider], outputs=[chatbox, response_text, input_text])
    regenerate_button.click(regenerate_text, inputs=[input_text, chatbox, output_length_slider, temperature_slider, top_k_slider, top_p_slider], outputs=response_text)

# Launch the interface
demo.launch()