update get rpm from api.openai.com

config/config2.yaml.example

@@ -3,6 +3,7 @@ llm:
   base_url: "YOUR_BASE_URL"
   api_key: "YOUR_API_KEY"
   model: "gpt-4-turbo-preview"  # or gpt-3.5-turbo-1106 / gpt-4-1106-preview
+  openai_session_key: ""  # If you want to automate the setup for RPM and MAX_TOKENS, please log in through your own browser at https://platform.openai.com/account/limits, enable developer mode, and find 'rate_limits' in the network section. If it's not found, refresh the recording using Ctrl+R. After locating 'rate_limits', find 'Authorization: sess-xxx' in the request header, where 'sess-xxx' is the session key.If you wish to manually set RPM, please set RPM and either do not set OPENAI_SESSION_KEY or set it to "".
 
 proxy: "YOUR_PROXY"
 

metagpt/configs/llm_config.py

@@ -40,6 +40,7 @@ class LLMConfig(YamlModel):
     api_type: LLMType = LLMType.OPENAI
     base_url: str = "https://api.openai.com/v1"
     api_version: Optional[str] = None
+    openai_session_key: Optional[str] = None
 
     model: Optional[str] = None  # also stands for DEPLOYMENT_NAME
 
@@ -63,6 +64,7 @@ class LLMConfig(YamlModel):
     logprobs: Optional[bool] = None  # https://cookbook.openai.com/examples/using_logprobs
     top_logprobs: Optional[int] = None
     timeout: int = 60
+    rpm: int = 10
 
     # For Network
     proxy: Optional[str] = None

metagpt/provider/openai_api.py

@@ -9,6 +9,9 @@
 
 import json
 from typing import AsyncIterator, Optional, Union
+from typing import AsyncIterator, Union
+
+import openai
 
 from openai import APIConnectionError, AsyncOpenAI, AsyncStream
 from openai._base_client import AsyncHttpxClientWrapper
@@ -30,6 +33,8 @@
 from metagpt.schema import Message
 from metagpt.utils.common import CodeParser, decode_image
 from metagpt.utils.cost_manager import CostManager, Costs
+from metagpt.utils.ahttp_client import aget
+from metagpt.utils.cost_manager import Costs
 from metagpt.utils.exceptions import handle_exception
 from metagpt.utils.token_counter import (
     count_message_tokens,
@@ -61,15 +66,15 @@ def __init__(self, config: LLMConfig):
         self.cost_manager: Optional[CostManager] = None
 
     def _init_model(self):
-        self.model = self.config.model  # Used in _calc_usage & _cons_kwargs
+        self.model = self.config.llm.model  # Used in _calc_usage & _cons_kwargs
 
     def _init_client(self):
         """https://github.com/openai/openai-python#async-usage"""
         kwargs = self._make_client_kwargs()
         self.aclient = AsyncOpenAI(**kwargs)
 
     def _make_client_kwargs(self) -> dict:
-        kwargs = {"api_key": self.config.api_key, "base_url": self.config.base_url}
+        kwargs = {"api_key": self.config.llm.api_key, "base_url": self.config.llm.base_url}
 
         # to use proxy, openai v1 needs http_client
         if proxy_params := self._get_proxy_params():
@@ -290,3 +295,54 @@ async def gen_image(
             img_url_or_b64 = item.url if resp_format == "url" else item.b64_json
             imgs.append(decode_image(img_url_or_b64))
         return imgs
+
+    async def update_rpm(self):
+        """
+        Asynchronously updates the RPM (requests per minute) limit.
+
+        This method fetches the RPM limit from an external API and updates the rate limiting parameters.
+        It is designed to be run before making any batched API calls.
+        """
+        self.rpm = await self._aget_rpm()
+        self.interval = 1.1 * 60 / self.rpm
+        logger.info(f'Setting rpm to {self.rpm}')
+
+    async def _aget_rpm(self) -> int:
+        """
+        Asynchronously fetches the RPM (requests per minute) limit from an external API.
+
+        This is an internal method used by update_rpm to fetch the current RPM limit. It uses
+        the OPENAI_SESSION_KEY for authorization and falls back to a default RPM value in case of failure.
+
+        Returns:
+            int: The fetched or default RPM value.
+        """
+        session_key = self.config.llm.openai_session_key
+        default_rpm = self.config.llm.rpm
+        if len(session_key) > 0:
+            try:
+                response = await aget(
+                                      url="https://api.openai.com/dashboard/rate_limits",
+                                      headers={"Authorization": f"Bearer {session_key}"},
+                                      proxy=self.config.llm.proxy
+                                     )
+                response_content = json.loads(response)
+                if not "error" in response_content:
+                    if self.model not in response_content:
+                        raise ValueError("Get rpm from api.openai.com error. \
+                                            You have entered a model name that is not supported by OpenAI, or the input is incorrect. \
+                                            Please enter the correct name in the configuration file. \
+                                            Setting rpm to default parameter.")
+
+                    limit_dict = response_content[self.model]
+                    return limit_dict["max_requests_per_1_minute"]
+                else:
+                    error = response_content["error"]
+                    logger.error(f"Connection to api.openai.com failed:{error}.Setting rpm to default parameter.")
+                    return default_rpm
+
+            except Exception as exp:
+                logger.error(f"Connection to api.openai.com failed, error type:{type(exp).__name__}, error message:{str(exp)}.Setting rpm to default parameter.")
+                return default_rpm
+        else:
+            return default_rpm

metagpt/utils/ahttp_client.py

@@ -28,6 +28,25 @@ async def apost(
     return data
 
 
+async def aget(
+    url: str,
+    params: Optional[Mapping[str, str]] = None,
+    headers: Optional[dict] = None,
+    as_json: bool = False,
+    encoding: str = "utf-8",
+    timeout: int = DEFAULT_TIMEOUT.total,
+    proxy: str = None,
+) -> Union[str, dict]:
+    async with aiohttp.ClientSession() as session:
+        async with session.get(url=url, params=params, headers=headers, timeout=timeout, proxy=proxy) as resp:
+            if as_json:
+                data = await resp.json()
+            else:
+                data = await resp.read()
+                data = data.decode(encoding)
+    return data
+
+
 async def apost_stream(
     url: str,
     params: Optional[Mapping[str, str]] = None,

tests/metagpt/provider/test_openai_api.py

@@ -0,0 +1,19 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+"""
+@Time    : 2023/12/18 19:08
+@Author  : zhongyang
+@File    : test_openai_api.py
+"""
+
+import pytest
+
+from metagpt.logs import logger
+from metagpt.provider.openai_api import OpenAILLM
+from metagpt.config2 import config
+
+@pytest.mark.asyncio
+async def test_update_rpm():
+    llm = OpenAILLM(config)
+    await llm.update_rpm()
+    assert isinstance(llm.rpm, float)