app.py

import gradio as gr
from transformers import pipeline

# Load models
model_1 = pipeline("text-generation", model="gpt2")
model_2 = pipeline("text-generation", model="tiiuae/falcon-rw-1b", trust_remote_code=True)
model_3 = pipeline("text2text-generation", model="google/flan-t5-small")

# Inference function
def compare_outputs(prompt):
    out1 = model_1(prompt, max_length=50, do_sample=True, temperature=0.7)[0]["generated_text"]
    out2 = model_2(prompt, max_length=50, do_sample=True, temperature=0.7)[0]["generated_text"]
    out3 = model_3(prompt, max_length=50)[0]["generated_text"]
    return out1.strip(), out2.strip(), out3.strip()

# Gradio interface
gr.Interface(
    fn=compare_outputs,
    inputs=gr.Textbox(lines=4, label="Your Prompt"),
    outputs=[
        gr.Textbox(label="GPT-2 Output"),
        gr.Textbox(label="Falcon-RW-1B Output"),
        gr.Textbox(label="FLAN-T5 Small Output"),
    ],
    title="🧪 LLM Prompt Behavior Explorer",
    description="Compare how small, open-source language models respond to the same prompt."
).launch()