aphilippov/python-server-api

1import re2import os3from pydantic import BaseModel, Extra, root_validator4from typing import Any, Callable, Dict, List, Optional, Union, Tuple5from time import sleep6 7from autogen._pydantic import PYDANTIC_V18from autogen.agentchat import Agent, UserProxyAgent9from autogen.code_utils import UNKNOWN, extract_code, execute_code, infer_lang10from autogen.math_utils import get_answer11 12 13PROMPTS = {14    # default15    "default": """Let's use Python to solve a math problem.16 17Query requirements:18You should always use the 'print' function for the output and use fractions/radical forms instead of decimals.19You can use packages like sympy to help you.20You must follow the formats below to write your code:21```python22# your code23```24 25First state the key idea to solve the problem. You may choose from three ways to solve the problem:26Case 1: If the problem can be solved with Python code directly, please write a program to solve it. You can enumerate all possible arrangements if needed.27Case 2: If the problem is mostly reasoning, you can solve it by yourself directly.28Case 3: If the problem cannot be handled in the above two ways, please follow this process:291. Solve the problem step by step (do not over-divide the steps).302. Take out any queries that can be asked through Python (for example, any calculations or equations that can be calculated).313. Wait for me to give the results.324. Continue if you think the result is correct. If the result is invalid or unexpected, please correct your query or reasoning.33 34After all the queries are run and you get the answer, put the answer in \\boxed{}.35 36Problem:37""",38    # select python or wolfram39    "two_tools": """Let's use two tools (Python and Wolfram alpha) to solve a math problem.40 41Query requirements:42You must follow the formats below to write your query:43For Wolfram Alpha:44```wolfram45# one wolfram query46```47For Python:48```python49# your code50```51When using Python, you should always use the 'print' function for the output and use fractions/radical forms instead of decimals. You can use packages like sympy to help you.52When using wolfram, give one query in each code block.53 54Please follow this process:551. Solve the problem step by step (do not over-divide the steps).562. Take out any queries that can be asked through Python or Wolfram Alpha, select the most suitable tool to be used (for example, any calculations or equations that can be calculated).573. Wait for me to give the results.584. Continue if you think the result is correct. If the result is invalid or unexpected, please correct your query or reasoning.59 60After all the queries are run and you get the answer, put the final answer in \\boxed{}.61 62Problem: """,63    # use python step by step64    "python": """Let's use Python to solve a math problem.65 66Query requirements:67You should always use the 'print' function for the output and use fractions/radical forms instead of decimals.68You can use packages like sympy to help you.69You must follow the formats below to write your code:70```python71# your code72```73 74Please follow this process:751. Solve the problem step by step (do not over-divide the steps).762. Take out any queries that can be asked through Python (for example, any calculations or equations that can be calculated).773. Wait for me to give the results.784. Continue if you think the result is correct. If the result is invalid or unexpected, please correct your query or reasoning.79 80After all the queries are run and you get the answer, put the answer in \\boxed{}.81 82Problem: """,83}84 85 86def _is_termination_msg_mathchat(message):87    """Check if a message is a termination message."""88    if isinstance(message, dict):89        message = message.get("content")90        if message is None:91            return False92    cb = extract_code(message)93    contain_code = False94    for c in cb:95        if c[0] == "python" or c[0] == "wolfram":96            contain_code = True97            break98    return not contain_code and get_answer(message) is not None and get_answer(message) != ""99 100 101def _add_print_to_last_line(code):102    """Add print() to the last line of a string."""103    # 1. check if there is already a print statement104    if "print(" in code:105        return code106    # 2. extract the last line, enclose it in print() and return the new string107    lines = code.splitlines()108    last_line = lines[-1]109    if "\t" in last_line or "=" in last_line:110        return code111    if "=" in last_line:112        last_line = "print(" + last_line.split(" = ")[0] + ")"113        lines.append(last_line)114    else:115        lines[-1] = "print(" + last_line + ")"116    # 3. join the lines back together117    return "\n".join(lines)118 119 120def _remove_print(code):121    """remove all print statements from a string."""122    lines = code.splitlines()123    lines = [line for line in lines if not line.startswith("print(")]124    return "\n".join(lines)125 126 127class MathUserProxyAgent(UserProxyAgent):128    """(Experimental) A MathChat agent that can handle math problems."""129 130    MAX_CONSECUTIVE_AUTO_REPLY = 15  # maximum number of consecutive auto replies (subject to future change)131    DEFAULT_REPLY = "Continue. Please keep solving the problem until you need to query. (If you get to the answer, put it in \\boxed{}.)"132 133    def __init__(134        self,135        name: Optional[str] = "MathChatAgent",  # default set to MathChatAgent136        is_termination_msg: Optional[137            Callable[[Dict], bool]138        ] = _is_termination_msg_mathchat,  # terminate if \boxed{} in message139        human_input_mode: Optional[str] = "NEVER",  # Fully automated140        default_auto_reply: Optional[Union[str, Dict, None]] = DEFAULT_REPLY,141        max_invalid_q_per_step=3,  # a parameter needed in MathChat142        **kwargs,143    ):144        """145        Args:146            name (str): name of the agent147            is_termination_msg (function): a function that takes a message in the form of a dictionary and returns a boolean value indicating if this received message is a termination message.148                The dict can contain the following keys: "content", "role", "name", "function_call".149            human_input_mode (str): whether to ask for human inputs every time a message is received.150                Possible values are "ALWAYS", "TERMINATE", "NEVER".151                (1) When "ALWAYS", the agent prompts for human input every time a message is received.152                    Under this mode, the conversation stops when the human input is "exit",153                    or when is_termination_msg is True and there is no human input.154                (2) When "TERMINATE", the agent only prompts for human input only when a termination message is received or155                    the number of auto reply reaches the max_consecutive_auto_reply.156                (3) (Default) When "NEVER", the agent will never prompt for human input. Under this mode, the conversation stops157                    when the number of auto reply reaches the max_consecutive_auto_reply or when is_termination_msg is True.158            default_auto_reply (str or dict or None): the default auto reply message when no code execution or llm based reply is generated.159            max_invalid_q_per_step (int): (ADDED) the maximum number of invalid queries per step.160            **kwargs (dict): other kwargs in [UserProxyAgent](../user_proxy_agent#__init__).161        """162        super().__init__(163            name=name,164            is_termination_msg=is_termination_msg,165            human_input_mode=human_input_mode,166            default_auto_reply=default_auto_reply,167            **kwargs,168        )169        self.register_reply([Agent, None], MathUserProxyAgent._generate_math_reply, position=2)170        # fixed var171        self._max_invalid_q_per_step = max_invalid_q_per_step172 173        # mutable174        self._valid_q_count = 0175        self._total_q_count = 0176        self._accum_invalid_q_per_step = 0177        self._previous_code = ""178        self.last_reply = None179 180    def generate_init_message(self, problem, prompt_type="default", customized_prompt=None):181        """Generate a prompt for the assistant agent with the given problem and prompt.182 183        Args:184            problem (str): the problem to be solved.185            prompt_type (str): the type of the prompt. Possible values are "default", "python", "wolfram".186                (1) "default": the prompt that allows the agent to choose between 3 ways to solve a problem:187                    1. write a python program to solve it directly.188                    2. solve it directly without python.189                    3. solve it step by step with python.190                (2) "python":191                    a simplified prompt from the third way of the "default" prompt, that asks the assistant192                    to solve the problem step by step with python.193                (3) "two_tools":194                    a simplified prompt similar to the "python" prompt, but allows the model to choose between195                    Python and Wolfram Alpha to solve the problem.196            customized_prompt (str): a customized prompt to be used. If it is not None, the prompt_type will be ignored.197 198        Returns:199            str: the generated prompt ready to be sent to the assistant agent.200        """201        self._reset()202        if customized_prompt is not None:203            return customized_prompt + problem204        return PROMPTS[prompt_type] + problem205 206    def _reset(self):207        # super().reset()208        self._valid_q_count = 0209        self._total_q_count = 0210        self._accum_invalid_q_per_step = 0211        self._previous_code = ""212        self.last_reply = None213 214    def execute_one_python_code(self, pycode):215        """Execute python code blocks.216 217        Previous python code will be saved and executed together with the new code.218        the "print" function will also be added to the last line of the code if needed219        """220        # Need to replace all "; " with "\n" to avoid syntax error when adding `print` to the last line221        pycode = pycode.replace("; ", "\n").replace(";", "\n")222        pycode = self._previous_code + _add_print_to_last_line(pycode)223 224        return_code, output, _ = execute_code(pycode, **self._code_execution_config, timeout=5)225        is_success = return_code == 0226 227        if not is_success:228            # Remove the file information from the error string229            pattern = r'File "/[^"]+\.py", line \d+, in .+\n'230            if isinstance(output, str):231                output = re.sub(pattern, "", output)232            output = "Error: " + output233        elif output == "":234            # Check if there is any print statement235            if "print" not in pycode:236                output = "No output found. Make sure you print the results."237                is_success = False238            else:239                output = "No output found."240                is_success = True241 242        if len(output) > 2000:243            output = "Your requested query response is too long. You might have made a mistake. Please revise your reasoning and query."244            is_success = False245 246        if is_success:247            # remove print and check if it still works248            tmp = self._previous_code + "\n" + _remove_print(pycode) + "\n"249            rcode, _, _ = execute_code(tmp, **self._code_execution_config)250        else:251            # only add imports and check if it works252            tmp = self._previous_code + "\n"253            for line in pycode.split("\n"):254                if "import" in line:255                    tmp += line + "\n"256            rcode, _, _ = execute_code(tmp, **self._code_execution_config)257 258        if rcode == 0:259            self._previous_code = tmp260        return output, is_success261 262    def execute_one_wolfram_query(self, query: str):263        """Run one wolfram query and return the output.264 265        Args:266            query: string of the query.267 268        Returns:269            output: string with the output of the query.270            is_success: boolean indicating whether the query was successful.271        """272        # wolfram query handler273        wolfram = WolframAlphaAPIWrapper()274        output, is_success = wolfram.run(query)275        if output == "":276            output = "Error: The wolfram query is invalid."277            is_success = False278        return output, is_success279 280    def _generate_math_reply(281        self,282        messages: Optional[List[Dict]] = None,283        sender: Optional[Agent] = None,284        config: Optional[Any] = None,285    ):286        """Generate an auto reply."""287        if messages is None:288            messages = self._oai_messages[sender]289        message = messages[-1]290        message = message.get("content", "")291        code_blocks = extract_code(message)292 293        if len(code_blocks) == 1 and code_blocks[0][0] == UNKNOWN:294            # no code block is found, lang should be `UNKNOWN``295            return True, self._default_auto_reply296        is_success, all_success = True, True297        reply = ""298        for code_block in code_blocks:299            lang, code = code_block300            if not lang:301                lang = infer_lang(code)302            if lang == "python":303                output, is_success = self.execute_one_python_code(code)304            elif lang == "wolfram":305                output, is_success = self.execute_one_wolfram_query(code)306            else:307                output = "Error: Unknown language."308                is_success = False309 310            reply += output + "\n"311            if not is_success:312                all_success = False313                self._valid_q_count -= 1  # count invalid queries314 315        reply = reply.strip()316 317        if self.last_reply == reply:318            return True, reply + "\nYour query or result is same from the last, please try a new approach."319        self.last_reply = reply320 321        if not all_success:322            self._accum_invalid_q_per_step += 1323            if self._accum_invalid_q_per_step > self._max_invalid_q_per_step:324                self._accum_invalid_q_per_step = 0325                reply = "Please revisit the problem statement and your reasoning. If you think this step is correct, solve it yourself and continue the next step. Otherwise, correct this step."326 327        return True, reply328 329 330# Modified based on langchain. Langchain is licensed under MIT License:331# The MIT License332 333# Copyright (c) Harrison Chase334 335# Permission is hereby granted, free of charge, to any person obtaining a copy336# of this software and associated documentation files (the "Software"), to deal337# in the Software without restriction, including without limitation the rights338# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell339# copies of the Software, and to permit persons to whom the Software is340# furnished to do so, subject to the following conditions:341 342# The above copyright notice and this permission notice shall be included in343# all copies or substantial portions of the Software.344 345# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR346# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,347# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE348# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER349# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,350# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN351# THE SOFTWARE.352 353 354def get_from_dict_or_env(data: Dict[str, Any], key: str, env_key: str, default: Optional[str] = None) -> str:355    """Get a value from a dictionary or an environment variable."""356    if key in data and data[key]:357        return data[key]358    elif env_key in os.environ and os.environ[env_key]:359        return os.environ[env_key]360    elif default is not None:361        return default362    else:363        raise ValueError(364            f"Did not find {key}, please add an environment variable"365            f" `{env_key}` which contains it, or pass"366            f"  `{key}` as a named parameter."367        )368 369 370class WolframAlphaAPIWrapper(BaseModel):371    """Wrapper for Wolfram Alpha.372 373    Docs for using:374 375    1. Go to wolfram alpha and sign up for a developer account376    2. Create an app and get your APP ID377    3. Save your APP ID into WOLFRAM_ALPHA_APPID env variable378    4. pip install wolframalpha379 380    """381 382    wolfram_client: Any  #: :meta private:383    wolfram_alpha_appid: Optional[str] = None384 385    class Config:386        """Configuration for this pydantic object."""387 388        if PYDANTIC_V1:389            extra = Extra.forbid390 391    @root_validator(skip_on_failure=True)392    def validate_environment(cls, values: Dict) -> Dict:393        """Validate that api key and python package exists in environment."""394        wolfram_alpha_appid = get_from_dict_or_env(values, "wolfram_alpha_appid", "WOLFRAM_ALPHA_APPID")395        values["wolfram_alpha_appid"] = wolfram_alpha_appid396 397        try:398            import wolframalpha399 400        except ImportError as e:401            raise ImportError("wolframalpha is not installed. Please install it with `pip install wolframalpha`") from e402        client = wolframalpha.Client(wolfram_alpha_appid)403        values["wolfram_client"] = client404 405        return values406 407    def run(self, query: str) -> Tuple[str, bool]:408        """Run query through WolframAlpha and parse result."""409        from urllib.error import HTTPError410 411        is_success = False  # added412        res = None413        for _ in range(20):414            try:415                res = self.wolfram_client.query(query)416                break417            except HTTPError:418                sleep(1)419            except Exception:420                return (421                    "Wolfram Alpha wasn't able to answer it. Please try a new query for wolfram or use python.",422                    is_success,423                )424        if res is None:425            return (426                "Wolfram Alpha wasn't able to answer it (may due to web error), you can try again or use python.",427                is_success,428            )429 430        try:431            if not res["@success"]:432                return (433                    "Your Wolfram query is invalid. Please try a new query for wolfram or use python.",434                    is_success,435                )436            assumption = next(res.pods).text437            answer = ""438            for result in res["pod"]:439                if result["@title"] == "Solution":440                    answer = result["subpod"]["plaintext"]441                if result["@title"] == "Results" or result["@title"] == "Solutions":442                    for i, sub in enumerate(result["subpod"]):443                        answer += f"ans {i}: " + sub["plaintext"] + "\n"444                    break445            if answer == "":446                answer = next(res.results).text447 448        except Exception:449            return (450                "Wolfram Alpha wasn't able to answer it. Please try a new query for wolfram or use python.",451                is_success,452            )453 454        if answer is None or answer == "":455            # We don't want to return the assumption alone if answer is empty456            return "No good Wolfram Alpha Result was found", is_success457        is_success = True458        return f"Assumption: {assumption} \nAnswer: {answer}", is_success459