OmniParser/demo/gradio/computer_use_demo/loop.py

"""
Agentic sampling loop that calls the Anthropic API and local implenmentation of anthropic-defined computer use tools.
"""
import time
import json
import asyncio
import platform
from collections.abc import Callable
from datetime import datetime
from enum import StrEnum
from typing import Any, cast, Dict

from anthropic import Anthropic, AnthropicBedrock, AnthropicVertex, APIResponse
from anthropic.types import (
    ToolResultBlockParam,
    TextBlock,
)
from anthropic.types.beta import (
    BetaContentBlock,
    BetaContentBlockParam,
    BetaImageBlockParam,
    BetaMessage,
    BetaMessageParam,
    BetaTextBlockParam,
    BetaToolResultBlockParam,
)
from computer_use_demo.tools import ToolResult

import torch

from computer_use_demo.gui_agent.anthropic_agent import AnthropicActor
from computer_use_demo.executor.anthropic_executor import AnthropicExecutor
from computer_use_demo.omniparser_agent.vlm_agent import OmniParser, VLMAgent
from computer_use_demo.colorful_text import colorful_text_vlm
from computer_use_demo.tools.screen_capture import get_screenshot
from computer_use_demo.gui_agent.llm_utils.oai import encode_image


BETA_FLAG = "computer-use-2024-10-22"


class APIProvider(StrEnum):
    ANTHROPIC = "anthropic"
    BEDROCK = "bedrock"
    VERTEX = "vertex"
    OPENAI = "openai"
    QWEN = "qwen"


PROVIDER_TO_DEFAULT_MODEL_NAME: dict[APIProvider, str] = {
    APIProvider.ANTHROPIC: "claude-3-5-sonnet-20241022",
    APIProvider.BEDROCK: "anthropic.claude-3-5-sonnet-20241022-v2:0",
    APIProvider.VERTEX: "claude-3-5-sonnet-v2@20241022",
    # APIProvider.OPENAI: "gpt-4o",
    # APIProvider.QWEN: "qwen2vl",
}


# This system prompt is optimized for the Docker environment in this repository and
# specific tool combinations enabled.
# We encourage modifying this system prompt to ensure the model has context for the
# environment it is running in, and to provide any additional information that may be
# helpful for the task at hand.
SYSTEM_PROMPT = f"""<SYSTEM_CAPABILITY>
* You are utilizing a Windows system with internet access.
* The current date is {datetime.today().strftime('%A, %B %d, %Y')}.
</SYSTEM_CAPABILITY>
"""

import base64
from PIL import Image
from io import BytesIO

def sampling_loop_sync(
    *,
    model: str,
    provider: APIProvider | None,
    messages: list[BetaMessageParam],
    output_callback: Callable[[BetaContentBlock], None],
    tool_output_callback: Callable[[ToolResult, str], None],
    api_response_callback: Callable[[APIResponse[BetaMessage]], None],
    api_key: str,
    only_n_most_recent_images: int | None = 2,
    max_tokens: int = 4096,
    selected_screen: int = 0
):
    """
    Synchronous agentic sampling loop for the assistant/tool interaction of computer use.
    """
    print('in sampling_loop_sync, model:', model)
    if model == "claude-3-5-sonnet-20241022":
        omniparser = OmniParser(url="http://localhost:8000/send_text/",
                                selected_screen=selected_screen,)
        
        # Register Actor and Executor
        actor = AnthropicActor(
            model=model, 
            provider=provider,
            api_key=api_key, 
            api_response_callback=api_response_callback,
            max_tokens=max_tokens,
            only_n_most_recent_images=only_n_most_recent_images,
            selected_screen=selected_screen
        )

        # from IPython.core.debugger import Pdb; Pdb().set_trace()
        executor = AnthropicExecutor(
            output_callback=output_callback,
            tool_output_callback=tool_output_callback,
            selected_screen=selected_screen
        )
    
    elif model == "omniparser + gpt-4o" or model == "omniparser + phi35v":
        # omniparser = OmniParser(url="http://localhost:8000/send_text/",
        #                         selected_screen=selected_screen,)
        omniparser = OmniParser(url=None,
                                selected_screen=selected_screen,)

        actor = VLMAgent(
            model=model,
            provider=provider,
            api_key=api_key,
            api_response_callback=api_response_callback,
            selected_screen=selected_screen,
            output_callback=output_callback,
        )

        executor = AnthropicExecutor(
            output_callback=output_callback,
            tool_output_callback=tool_output_callback,
            selected_screen=selected_screen
        )
        
    else:
        raise ValueError(f"Model {model} not supported")
    print(f"Model Inited: {model}, Provider: {provider}")
    
    tool_result_content = None
    
    print(f"Start the message loop. User messages: {messages}")
    
    if model == "claude-3-5-sonnet-20241022": # Anthropic loop
        while True:
            parsed_screen = omniparser() # parsed_screen: {"som_image_base64": dino_labled_img, "parsed_content_list": parsed_content_list, "screen_info"}
            import pdb; pdb.set_trace()
            screen_info_block = TextBlock(text='Below is the structured accessibility information of the current UI screen, which includes text and icons you can operate on, take these information into account when you are making the prediction for the next action. Note you will still need to take screenshot to get the image: \n' + parsed_screen['screen_info'], type='text')
            # # messages[-1]['content'].append(screen_info_block)
            screen_info_dict = {"role": "user", "content": [screen_info_block]}
            messages.append(screen_info_dict)
            tools_use_needed = actor(messages=messages)

            for message, tool_result_content in executor(tools_use_needed, messages):
                yield message
        
            if not tool_result_content:
                return messages

            messages.append({"content": tool_result_content, "role": "user"})
    
    elif model == "omniparser + gpt-4o" or model == "omniparser + phi35v":
        while True:
            parsed_screen = omniparser()
            tools_use_needed, vlm_response_json = actor(messages=messages, parsed_screen=parsed_screen)

            for message, tool_result_content in executor(tools_use_needed, messages):
                yield message
        
            if not tool_result_content:
                return messages
            
            # import pdb; pdb.set_trace()
            # messages.append({"role": "user",
            #                  "content": ["History plan:\n" + str(vlm_response_json['Reasoning'])]})

            # messages.append({"content": tool_result_content, "role": "user"})