app.py

import gradio as gr
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

# Load the models
model_path_1 = "ibm-granite/granite-3.0-1b-a400m-instruct"
model_path_2 = "ibm-granite/granite-3.0-1b-a400m-base"

tokenizer_1 = AutoTokenizer.from_pretrained(model_path_1)
tokenizer_2 = AutoTokenizer.from_pretrained(model_path_2)

model_1 = AutoModelForCausalLM.from_pretrained(model_path_1, device_map="auto")
model_2 = AutoModelForCausalLM.from_pretrained(model_path_2, device_map="auto")

model_1.eval()
model_2.eval()

# Mood prompts dictionary
mood_prompts = {
    "Fun": "Respond in a light-hearted, playful manner.",
    "Serious": "Respond in a thoughtful, serious tone.",
    "Professional": "Respond in a formal, professional manner.",
    "Upset": "Respond in a slightly irritated, upset tone.",
    "Empathetic": "Respond in a warm and understanding tone.",
    "Optimistic": "Respond in a positive, hopeful manner.",
    "Sarcastic": "Respond with a hint of sarcasm.",
    "Motivational": "Respond with encouragement and motivation.",
    "Curious": "Respond with a sense of wonder and curiosity.",
    "Humorous": "Respond with a touch of humor.",
    "Cautious": "Respond with careful consideration and caution.",
    "Assertive": "Respond with confidence and assertiveness.",
    "Friendly": "Respond in a warm and friendly manner.",
    "Romantic": "Respond with affection and romance.",
    "Nostalgic": "Respond with a sense of longing for the past.",
    "Grateful": "Respond with gratitude and appreciation.",
    "Inspirational": "Respond with inspiration and positivity.",
    "Casual": "Respond in a relaxed and informal tone.",
    "Formal": "Respond with a high level of formality.",
    "Pessimistic": "Respond with a focus on potential negatives.",
    "Excited": "Respond with enthusiasm and excitement.",
    "Melancholic": "Respond with a sense of sadness or longing.",
    "Confident": "Respond with self-assurance and confidence.",
    "Suspicious": "Respond with caution and doubt.",
    "Reflective": "Respond with deep thought and introspection.",
    "Joyful": "Respond with happiness and joy.",
    "Mysterious": "Respond with an air of mystery and intrigue.",
    "Aggressive": "Respond with force and intensity.",
    "Calm": "Respond with a sense of peace and tranquility.",
    "Gloomy": "Respond with a sense of sadness or pessimism.",
    "Encouraging": "Respond with words of support and encouragement.",
    "Sympathetic": "Respond with understanding and compassion.",
    "Disappointed": "Respond with a tone of disappointment.",
    "Proud": "Respond with a sense of pride and accomplishment.",
    "Playful": "Respond in a fun and playful manner.",
    "Inquisitive": "Respond with curiosity and interest.",
    "Supportive": "Respond with reassurance and support.",
    "Reluctant": "Respond with hesitation and reluctance.",
    "Confused": "Respond with uncertainty and confusion.",
    "Energetic": "Respond with high energy and enthusiasm.",
    "Relaxed": "Respond with a calm and laid-back tone.",
    "Grumpy": "Respond with a touch of irritation.",
    "Hopeful": "Respond with a sense of hope and optimism.",
    "Indifferent": "Respond with a lack of strong emotion.",
    "Surprised": "Respond with shock and astonishment.",
    "Tense": "Respond with a sense of urgency or anxiety.",
    "Enthusiastic": "Respond with eagerness and excitement.",
    "Worried": "Respond with concern and apprehension."
}


def generate_response(prompt, mood, max_new_tokens, temperature, top_p, repetition_penalty, model_choice):
    mood_prompt = mood_prompts.get(mood, "")
    full_prompt = f"{mood_prompt} {prompt}"
    
    # Choose model and tokenizer based on user selection
    if model_choice == "Granite 3.0-1B A400M Instruct":
        model = model_1
        tokenizer = tokenizer_1
    else:
        model = model_2
        tokenizer = tokenizer_2

    input_tokens = tokenizer(full_prompt, return_tensors="pt").to(model.device)
    
    output = model.generate(
        **input_tokens,
        max_new_tokens=max_new_tokens,
        temperature=temperature,
        top_p=top_p,
        repetition_penalty=repetition_penalty,
        do_sample=True
    )
    
    response = tokenizer.decode(output[0], skip_special_tokens=True)
    return response.strip()

with gr.Blocks(theme="prithivMLmods/Minecraft-Theme") as demo:
    
    with gr.Row():
        with gr.Column():
            prompt = gr.Textbox(label="Prompt", placeholder="Enter your prompt here...", lines=5)
            mood = gr.Dropdown(label="Select Mood", choices=list(mood_prompts.keys()), value="Professional")
            model_choice = gr.Radio(label="Select Model", choices=["Granite 3.0-1B A400M Instruct", "Granite 3.0-1B A400M Base"], value="Granite 3.0-1B A400M Instruct")
            generate_button = gr.Button("Generate Response")
            max_new_tokens = gr.Slider(minimum=1, maximum=500, value=100, step=1, label="Max New Tokens")
            temperature = gr.Slider(minimum=0.1, maximum=2.0, value=0.7, step=0.1, label="Temperature")
            top_p = gr.Slider(minimum=0.1, maximum=1.0, value=0.9, step=0.1, label="Top P")
            repetition_penalty = gr.Slider(minimum=1.0, maximum=2.0, value=1.1, step=0.1, label="Repetition Penalty")
    
        with gr.Column():
            output = gr.Textbox(label="Response", lines=15)
    
    generate_button.click(
        generate_response,
        inputs=[prompt, mood, max_new_tokens, temperature, top_p, repetition_penalty, model_choice],
        outputs=output
    )
    
    gr.Markdown("## Examples")
    examples = gr.Examples(
        examples=[
            ["Give me advice on staying motivated.", "Motivational", 100, 0.7, 0.9, 1.1, "Granite 3.0-1B A400M Instruct"],
            ["Describe a futuristic city.", "Optimistic", 200, 0.9, 0.8, 1.0, "Granite 3.0-1B A400M Base"]
        ],
        inputs=[prompt, mood, max_new_tokens, temperature, top_p, repetition_penalty, model_choice],
    )

demo.launch(share=True)