app.py

import os
import asyncio
from concurrent.futures import ThreadPoolExecutor
import requests
import gradio as gr

MAX_NEW_TOKENS = 256
TOKEN = os.environ.get("HF_TOKEN", None)
URLS = [
    "https://api-inference.huggingface.co/models/google/flan-ul2",
    "https://api-inference.huggingface.co/models/google/flan-t5-xxl",
]

def fetch(session, text, api_url):
    model = api_url.split("/")[-1]
    response = session.post(api_url, json={"inputs": text, "parameters": {"max_new_tokens": MAX_NEW_TOKENS}})
    if response.status_code != 200:
        return model, None
    return model, response.json()

async def inference(text):
    with ThreadPoolExecutor(max_workers=2) as executor:
        with requests.Session() as session:
            session.headers = {"Authorization": f"Bearer {TOKEN}"}
            # Initialize the event loop
            loop = asyncio.get_event_loop()
            tasks = [
                loop.run_in_executor(
                    executor, fetch, *(session, text, url)  # Allows us to pass in multiple arguments to `fetch`
                )
                for url in URLS
            ]

            # Initializes the tasks to run and awaits their results
            responses = [None, None]
            for (model, response) in await asyncio.gather(*tasks):
                if response is not None:
                    if model == "flan-ul2":
                        responses[0] = response[0]["generated_text"]
                    elif model == "flan-t5-xxl":
                        responses[1] = response[0]["generated_text"]
    return responses

def feedback(inputs, feedback, is_positive):
    with open('promptlog.txt', 'a') as f:
        f.write(f"Inputs: {inputs}\nFeedback: {feedback}\nIs positive: {is_positive}\n\n")

def display_history():
    try:
        with open('promptlog.txt', 'r') as f:
            history = f.read()
    except FileNotFoundError:
        history = "No history yet."
    print(history)

def app():
    title = "Flan UL2 vs Flan T5 XXL"
    description = "Compare with feedback: [Flan-T5-xxl](https://huggingface.co/google/flan-t5-xxl) and [Flan-UL2](https://huggingface.co/google/flan-ul2)."
    inputs = gr.inputs.Textbox(lines=3, label="Input Prompt")
    outputs = [gr.outputs.Textbox(lines=3, label="Flan T5-UL2"), gr.outputs.Textbox(lines=3, label="Flan T5-XXL")]
    feedback_box = gr.inputs.CheckboxGroup(["Positive feedback", "Negative feedback"], label="Feedback")
    feedback_text = gr.inputs.Textbox(label="Feedback Reason")
    feedback_button = gr.inputs.Button(label="Submit Feedback")
    display_history_button = gr.inputs.Button(label="Display Feedback History")

    def predict_text(inputs):
        return inference(inputs)

    def handle_feedback(inputs, feedback, is_positive):
        feedback(inputs, feedback, is_positive)
        return "Thank you for your feedback!"

    def handle_display_history():
        display_history()

    gr.Interface(fn=predict_text, inputs=inputs, outputs=outputs, title=title, description=description).launch()

    feedback_ui = gr.Interface(fn=handle_feedback, inputs=[inputs, feedback_box, feedback_text, feedback_button], outputs=gr.outputs.Textbox(label="Feedback Submitted"), title="Feedback", description="Please provide feedback on the model's response.")

display_history_ui = gr.Interface(fn=handle_display_history, inputs=display_history_button, outputs=gr.outputs.Textbox(label="Feedback History"), title="Feedback History", description="View history of feedback submissions.")

gr.Interface([feedback_ui, display_history_ui], columns=2, title="Flan Feedback").launch()

if name == 'main':
    app()