ControllerAtomicFlow.py

import importlib
import importlib.util
import json
import os.path
from copy import deepcopy
from typing import Any, Dict, List
from flow_modules.aiflows.OpenAIChatFlowModule import OpenAIChatAtomicFlow

from dataclasses import dataclass


@dataclass
class Command:
    name: str
    description: str
    input_args: List[str]


class ControllerAtomicFlow(OpenAIChatAtomicFlow):
    def __init__(
            self,
            commands: List[Command],
            plan_file_location: str,
            code_file_location: str,
            **kwargs):
        super().__init__(**kwargs)
        if os.path.isdir(plan_file_location):
            plan_file_location = os.path.join(plan_file_location, "plan.txt")
        self.plan_file_location = plan_file_location
        self.code_file_location = code_file_location
        self.system_message_prompt_template = self.system_message_prompt_template.partial(
            commands=self._build_commands_manual(commands)
        )
        self.hint_for_model = """
        Make sure your response is in the following format:
              Response Format:
              {
              "thought": "thought",
              "reasoning": "reasoning",
              "criticism": "constructive self-criticism",
              "speak": "thoughts summary to say to user",
              "command": "the python function you would like to call",
              "command_args": {
                  "arg name": "value"
                  }
              }
        """
        self.original_system_template = self.system_message_prompt_template.template

    @staticmethod
    def _build_commands_manual(commands: List[Command]) -> str:
        ret = ""
        for i, command in enumerate(commands):
            command_input_json_schema = json.dumps(
                {input_arg: f"YOUR_{input_arg.upper()}" for input_arg in command.input_args})
            ret += f"{i + 1}. {command.name}: {command.description} Input arguments (given in the JSON schema): {command_input_json_schema}\n"
        return ret

    @classmethod
    def instantiate_from_config(cls, config):
        flow_config = deepcopy(config)

        kwargs = {"flow_config": flow_config}

        # ~~~ Set up prompts ~~~
        kwargs.update(cls._set_up_prompts(flow_config))
        kwargs.update(cls._set_up_backend(flow_config))

        # ~~~ Set up commands ~~~
        commands = flow_config["commands"]
        commands = [
            Command(name, command_conf["description"], command_conf["input_args"]) for name, command_conf in
            commands.items()
        ]
        kwargs.update({"commands": commands})

        # ~~~ Instantiate flow ~~~
        return cls(**kwargs)

    def _get_library_function_signatures(self):
        try:
            spec = importlib.util.spec_from_file_location("code_library", self.code_file_location)
            module = importlib.util.module_from_spec(spec)
            spec.loader.exec_module(module)
            ret = ''
            import inspect
            for name, obj in inspect.getmembers(module):
                if inspect.isfunction(obj):
                    ret += f"{name}: {inspect.signature(obj)}\n"
            return ret
        except FileNotFoundError:
            return 'There is no function available yet.'

    def _get_plan(self):
        try:
            with open(self.plan_file_location, 'r') as file:
                return file.read()
        except FileNotFoundError:
            return "There is no plan yet"

    def run(self, input_data: Dict[str, Any]) -> Dict[str, Any]:
        if 'goal' in input_data:
            input_data['goal'] += self.hint_for_model
        if 'human_feedback' in input_data:
            input_data['human_feedback'] += self.hint_for_model

        # self.system_message_prompt_template.template
        plan_to_append = self._get_plan()
        function_signatures_to_append = self._get_library_function_signatures()
        self.system_message_prompt_template.template = \
            self.original_system_template + "\n" + f"Here are the available functions at {self.code_file_location}\n" \
            + function_signatures_to_append + "\n" \
            + f"Here is the step-by-step plan to achieve the goal:\n" \
            + plan_to_append + "\n"

        api_output = super().run(input_data)["api_output"].strip()

        try:
            response = json.loads(api_output)
            return response
        except json.decoder.JSONDecodeError:
            new_input_data = input_data.copy()
            new_input_data['observation'] = ""
            new_input_data['human_feedback'] = "The previous respond cannot be parsed with json.loads, it could be the backslashes used for escaping single quotes in the string arguments of the Python code are not properly escaped themselves within the JSON context. Make sure your next response is in JSON format."
            new_api_output = super().run(new_input_data)["api_output"].strip()
            return json.loads(new_api_output)