app.py

import os
import subprocess
from playwright.async_api import async_playwright
import gradio as gr
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline

# Define model and inference parameters
MODEL_NAME = "HuggingFaceH4/zephyr-7b-beta"
MAX_LENGTH = 512
NUM_BEAMS = 5
TEMPERATURE = 0.7

# Function to generate actions using Zephyr-7b-beta model
def generate_actions(input_text):
    tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
    model = AutoModelForCausalLM.from_pretrained(MODEL_NAME)

    # Use pipeline for text generation
    generator = pipeline('text-generation', model=model, tokenizer=tokenizer)
    outputs = generator(input_text, max_length=MAX_LENGTH, num_beams=NUM_BEAMS, temperature=TEMPERATURE)

    response = outputs[0]['generated_text']
    actions = response.split("\n")
    return actions

# Function to initialize browser and page (now asynchronous)
async def initialize_browser():
    # Attempt to download browser binaries if not already present
    try:
        subprocess.check_output(["which", "chromium"], stderr=subprocess.STDOUT)
    except subprocess.CalledProcessError:
        print("Chromium not found. Attempting to download...")
        # Note: Directly executing 'playwright install' in Python is not supported.
        # You would need to ensure the environment is pre-configured or use a different method to install the browser.

    async with async_playwright() as p:
        browser = await p.chromium.launch()
        page = await browser.new_page()
        return browser, page

# Gradio interface (now using the asynchronous function)
async def run_agent(input_text):
    async with async_playwright() as p:
        browser, page = await initialize_browser()
        actions = generate_actions(input_text)

        for action in actions:
            if "open website" in action:
                website = action.split(" ")[-1]
                await page.goto(website)
            elif "click" in action:
                selector = action.split(" ")[-1]
                await page.click(selector)
            elif "type" in action:
                text = action.split(" ")[-1]
                await page.type(text)
            elif "submit" in action:
                await page.press("Enter")
            else:
                print(f"Action not recognized: {action}")

        return f"Successfully executed actions based on: {input_text}"

iface = gr.Interface(
    fn=run_agent,  # Pass the async function directly
    inputs=gr.Textbox(label="Enter your request"),
    outputs=gr.Textbox(label="Response"),
    title="Automated Agent",
    description="Enter a task or instruction for the agent to perform."
)
iface.launch(share=True)  # Enable sharing