app.py

# ✅ Evo Showcase Mode: Gradio App with Enhanced Info Panel + GPT-3.5 Comparison

import gradio as gr
import openai
from inference import predict as evo_predict

# 🔐 SET YOUR GPT-3.5 API KEY HERE
openai.api_key = "sk-..."  # Replace with your actual key

# ✅ Use the new openai>=1.0.0 API
client = openai.OpenAI()

def gpt_predict(prompt):
    try:
        response = client.chat.completions.create(
            model="gpt-3.5-turbo",
            messages=[
                {"role": "system", "content": "You're a commonsense reasoning assistant. Only say: Solution 1 or Solution 2."},
                {"role": "user", "content": prompt}
            ]
        )
        return response.choices[0].message.content.strip()
    except Exception as e:
        return f"GPT Error: {str(e)}"

def compare(goal, sol1, sol2):
    if not goal.strip() or not sol1.strip() or not sol2.strip():
        return "⚠️ Please provide all inputs.", "", ""

    prompt = f"Goal: {goal}\nSolution 1: {sol1}\nSolution 2: {sol2}\nWhich is better?"
    evo = evo_predict(goal, sol1, sol2)
    gpt = gpt_predict(prompt)

    if evo == gpt:
        verdict = "✅ Evo agrees with GPT-3.5"
    else:
        verdict = "⚖️ Evo disagrees with GPT-3.5 — explore why."

    return f"🧠 Evo: {evo}", f"🤖 GPT-3.5: {gpt}", verdict

examples = [
    ["Start a fire", "Use a match", "Pour water"],
    ["Warm up food", "Use microwave", "Put it in fridge"],
    ["Charge a phone", "Plug it in", "Put it on grass"],
    ["Stop a car", "Press the brake", "Press the horn"]
]

with gr.Blocks(title="⚔️ Evo vs GPT-3.5 – Real-Time Commonsense Showdown") as demo:
    gr.Markdown("""
    # 🧠 EvoTransformer v2.1
    **PIQA Accuracy:** 69.7%     |     **Model Size:** ~13M Parameters     |     **Baseline:** GPT-3.5 ≈ 81%

    EvoTransformer is a scratch-built reasoning model trained on just 1K PIQA examples. No pretraining. No fine-tuning. Pure evolution.

    Compare its decisions with GPT-3.5 in real-time and witness how intelligence can emerge even from lean, efficient architectures.
    """)

    with gr.Row():
        goal = gr.Textbox(label="Goal")
    with gr.Row():
        sol1 = gr.Textbox(label="Solution 1")
        sol2 = gr.Textbox(label="Solution 2")

    evo_output = gr.Textbox(label="EvoTransformer Response")
    gpt_output = gr.Textbox(label="GPT-3.5 Response")
    verdict_output = gr.Textbox(label="Verdict")

    submit = gr.Button("Submit")
    submit.click(fn=compare, inputs=[goal, sol1, sol2], outputs=[evo_output, gpt_output, verdict_output])

    gr.Examples(examples=examples, inputs=[goal, sol1, sol2], outputs=[evo_output, gpt_output, verdict_output], fn=compare, cache_examples=False)

    gr.Markdown("""
    > 🧪 *Note: EvoTransformer is a scratch-built model trained on 1K PIQA examples. It may occasionally misinterpret context or idioms. That’s part of its evolution.*

    ---
    Made with ❤️ by Dr. Heman Mohabeer — EvoTransformer is not just code. It's evolution.
    """)

    demo.launch()