Merge remote-tracking branch 'upstream/master'

config.py

@@ -83,10 +83,12 @@
 
 # 模型选择是 (注意: LLM_MODEL是默认选中的模型, 它*必须*被包含在AVAIL_LLM_MODELS列表中 )
 LLM_MODEL = "gpt-3.5-turbo" # 可选 ↓↓↓
-AVAIL_LLM_MODELS = ["gpt-3.5-turbo-16k", "gpt-3.5-turbo", "azure-gpt-3.5", "api2d-gpt-3.5-turbo",
-                    "gpt-4", "gpt-4-32k", "azure-gpt-4", "api2d-gpt-4", "chatglm", "moss", "newbing", "stack-claude"]
+AVAIL_LLM_MODELS = ["gpt-3.5-turbo-16k", "gpt-3.5-turbo", "azure-gpt-3.5",
+                    "api2d-gpt-3.5-turbo", 'api2d-gpt-3.5-turbo-16k', 
+                    "gpt-4", "gpt-4-32k", "azure-gpt-4", "api2d-gpt-4", 
+                    "chatglm", "moss", "newbing", "claude-2"]
 # P.S. 其他可用的模型还包括 ["qianfan", "llama2", "qwen", "gpt-3.5-turbo-0613", "gpt-3.5-turbo-16k-0613",  "gpt-3.5-random"
-# "spark", "sparkv2", "chatglm_onnx", "claude-1-100k", "claude-2", "internlm", "jittorllms_pangualpha", "jittorllms_llama"]
+# "spark", "sparkv2", "sparkv3", "chatglm_onnx", "claude-1-100k", "claude-2", "internlm", "jittorllms_pangualpha", "jittorllms_llama"]
 
 
 # 百度千帆（LLM_MODEL="qianfan"）
@@ -139,12 +141,16 @@
 SLACK_CLAUDE_USER_TOKEN = ''
 
 
-# 如果需要使用AZURE 详情请见额外文档 docs\use_azure.md
+# 如果需要使用AZURE（方法一：单个azure模型部署）详情请见额外文档 docs\use_azure.md
 AZURE_ENDPOINT = "https://你亲手写的api名称.openai.azure.com/"
 AZURE_API_KEY = "填入azure openai api的密钥"    # 建议直接在API_KEY处填写，该选项即将被弃用
 AZURE_ENGINE = "填入你亲手写的部署名"            # 读 docs\use_azure.md
 
 
+# 如果需要使用AZURE（方法二：多个azure模型部署+动态切换）详情请见额外文档 docs\use_azure.md
+AZURE_CFG_ARRAY = {}
+
+
 # 使用Newbing (不推荐使用，未来将删除)
 NEWBING_STYLE = "creative"  # ["creative", "balanced", "precise"]
 NEWBING_COOKIES = """
@@ -191,6 +197,10 @@
 ALLOW_RESET_CONFIG = False
 
 
+# 在使用AutoGen插件时，是否使用Docker容器运行代码
+AUTOGEN_USE_DOCKER = True
+
+
 # 临时的上传文件夹位置，请勿修改
 PATH_PRIVATE_UPLOAD = "private_upload"
 
@@ -216,13 +226,16 @@
 │   ├── API_ORG（不常用）
 │   └── API_URL_REDIRECT（不常用）
 │
-├── "azure-gpt-3.5" 等azure模型
+├── "azure-gpt-3.5" 等azure模型（单个azure模型，不需要动态切换）
 │   ├── API_KEY
 │   ├── AZURE_ENDPOINT
 │   ├── AZURE_API_KEY
 │   ├── AZURE_ENGINE
 │   └── API_URL_REDIRECT
 │
+├── "azure-gpt-3.5" 等azure模型（多个azure模型，需要动态切换，高优先级）
+│   └── AZURE_CFG_ARRAY
+│
 ├── "spark" 星火认知大模型 spark & sparkv2
 │   ├── XFYUN_APPID
 │   ├── XFYUN_API_SECRET

crazy_functions/agent_fns/auto_agent.py

@@ -1,7 +1,7 @@
 from toolbox import CatchException, update_ui, gen_time_str, trimmed_format_exc, ProxyNetworkActivate
 from toolbox import report_execption, get_log_folder, update_ui_lastest_msg, Singleton
 from crazy_functions.agent_fns.pipe import PluginMultiprocessManager, PipeCom
-from crazy_functions.agent_fns.autogen_general import AutoGenGeneral
+from crazy_functions.agent_fns.general import AutoGenGeneral
 import time
 
 

crazy_functions/agent_fns/autogen_general.py → crazy_functions/agent_fns/general.py

@@ -36,12 +36,8 @@ def do_audogen(self, input):
         # ⭐⭐ 子进程执行
         input = input.content
         with ProxyNetworkActivate("AutoGen"):
-            from autogen import AssistantAgent, UserProxyAgent
-            config_list = [{
-                'model': self.llm_kwargs['llm_model'], 
-                'api_key': self.llm_kwargs['api_key'],
-            },]
-            code_execution_config={"work_dir": self.autogen_work_dir, "use_docker":True}
+            config_list = self.get_config_list()
+            code_execution_config={"work_dir": self.autogen_work_dir, "use_docker":self.use_docker}
             agents = self.define_agents()
             user_proxy = None
             assistant = None
@@ -67,6 +63,20 @@ def do_audogen(self, input):
                 tb_str = '```\n' + trimmed_format_exc() + '```'
                 self.child_conn.send(PipeCom("done", "AutoGen 执行失败: \n\n" + tb_str))
 
+    def get_config_list(self):
+        model = self.llm_kwargs['llm_model']
+        api_base = None
+        if self.llm_kwargs['llm_model'].startswith('api2d-'):
+            model = self.llm_kwargs['llm_model'][len('api2d-'):]
+            api_base = "https://openai.api2d.net/v1"
+        config_list = [{
+                'model': model, 
+                'api_key': self.llm_kwargs['api_key'],
+            },]
+        if api_base is not None:
+            config_list[0]['api_base'] = api_base
+        return config_list
+
     def subprocess_worker(self, child_conn):
         # ⭐⭐ 子进程执行
         self.child_conn = child_conn

crazy_functions/agent_fns/pipe.py

@@ -1,4 +1,5 @@
-from toolbox import get_log_folder, update_ui, gen_time_str, trimmed_format_exc, promote_file_to_downloadzone
+from toolbox import get_log_folder, update_ui, gen_time_str, get_conf, promote_file_to_downloadzone
+from crazy_functions.agent_fns.watchdog import WatchDog
 import time, os
 
 class PipeCom():
@@ -19,6 +20,16 @@ def __init__(self, llm_kwargs, plugin_kwargs, chatbot, history, system_prompt, w
         self.system_prompt = system_prompt
         self.web_port = web_port
         self.alive = True
+        self.use_docker, = get_conf('AUTOGEN_USE_DOCKER')
+
+        # create a thread to monitor self.heartbeat, terminate the instance if no heartbeat for a long time
+        timeout_seconds = 5*60
+        self.heartbeat_watchdog = WatchDog(timeout=timeout_seconds, bark_fn=self.terminate, interval=5)
+        self.heartbeat_watchdog.begin_watch()
+
+    def feed_heartbeat_watchdog(self):
+        # feed this `dog`, so the dog will not `bark` (bark_fn will terminate the instance)
+        self.heartbeat_watchdog.feed()
 
     def is_alive(self):
         return self.alive
@@ -50,7 +61,7 @@ def immediate_showoff_when_possible(self, fp):
         # 获取fp的拓展名
         file_type = fp.split('.')[-1]
         # 如果是文本文件, 则直接显示文本内容
-        if file_type in ['png', 'jpg']:
+        if file_type.lower() in ['png', 'jpg']:
             image_path = os.path.abspath(fp)
             self.chatbot.append(['检测到新生图像:', f'本地文件预览: <br/><div align="center"><img src="file={image_path}"></div>'])
             yield from update_ui(chatbot=self.chatbot, history=self.history)
@@ -79,7 +90,7 @@ def overwatch_workdir_file_change(self):
             for f in change_list: 
                 res = promote_file_to_downloadzone(f)
                 file_links += f'<br/><a href="file={res}" target="_blank">{res}</a>'
-                yield from self.immediate_showoff_when_possible(file_path)
+                yield from self.immediate_showoff_when_possible(f)
 
             self.chatbot.append(['检测到新生文档.', f'文档清单如下: {file_links}'])
             yield from update_ui(chatbot=self.chatbot, history=self.history)
@@ -98,9 +109,17 @@ def main_process_ui_control(self, txt, create_or_resume) -> str:
             self.terminate()
             return "terminate"
 
+        # patience = 10
+
         while True:
             time.sleep(0.5)
+            if not self.alive:
+                # the heartbeat watchdog might have it killed
+                self.terminate()
+                return "terminate"
+
             if self.parent_conn.poll(): 
+                self.feed_heartbeat_watchdog()
                 if '[GPT-Academic] 等待中' in self.chatbot[-1][-1]:
                     self.chatbot.pop(-1)    # remove the last line
                 msg = self.parent_conn.recv() # PipeCom
@@ -124,10 +143,17 @@ def main_process_ui_control(self, txt, create_or_resume) -> str:
                     # do not terminate here, leave the subprocess_worker instance alive
                     return "wait_feedback"
             else:
+                self.feed_heartbeat_watchdog()
                 if '[GPT-Academic] 等待中' not in self.chatbot[-1][-1]:
+                    # begin_waiting_time = time.time()
                     self.chatbot.append(["[GPT-Academic] 等待AutoGen执行结果 ...", "[GPT-Academic] 等待中"])
                 self.chatbot[-1] = [self.chatbot[-1][0], self.chatbot[-1][1].replace("[GPT-Academic] 等待中", "[GPT-Academic] 等待中.")]
                 yield from update_ui(chatbot=self.chatbot, history=self.history)
+                # if time.time() - begin_waiting_time > patience:
+                #     self.chatbot.append([f"结束", "等待超时, 终止AutoGen程序。"])
+                #     yield from update_ui(chatbot=self.chatbot, history=self.history)
+                #     self.terminate()
+                #     return "terminate"
 
         self.terminate()
         return "terminate"

crazy_functions/agent_fns/watchdog.py

@@ -0,0 +1,28 @@
+import threading, time
+
+class WatchDog():
+    def __init__(self, timeout, bark_fn, interval=3, msg="") -> None:
+        self.last_feed = None
+        self.timeout = timeout
+        self.bark_fn = bark_fn
+        self.interval = interval
+        self.msg = msg
+        self.kill_dog = False
+
+    def watch(self):
+        while True:
+            if self.kill_dog: break
+            if time.time() - self.last_feed > self.timeout:
+                if len(self.msg) > 0: print(self.msg)
+                self.bark_fn()
+                break
+            time.sleep(self.interval)
+
+    def begin_watch(self):
+        self.last_feed = time.time()
+        th = threading.Thread(target=self.watch)
+        th.daemon = True
+        th.start()
+
+    def feed(self):
+        self.last_feed = time.time()

crazy_functions/多智能体.py

@@ -1,10 +1,7 @@
 # 本源代码中, ⭐ = 关键步骤
 """
 测试：
-    - 裁剪图像，保留下半部分
-    - 交换图像的蓝色通道和红色通道
-    - 将图像转为灰度图像
-    - 将csv文件转excel表格
+    - show me the solution of $x^2=cos(x)$, solve this problem with figure, and plot and save image to t.jpg
 
 Testing: 
     - Crop the image, keeping the bottom half. 
@@ -35,7 +32,8 @@ def 多智能体终端(txt, llm_kwargs, plugin_kwargs, chatbot, history, system_
     web_port        当前软件运行的端口号
     """
     # 检查当前的模型是否符合要求
-    supported_llms = ['gpt-3.5-turbo-16k', 'gpt-4', 'gpt-4-32k']
+    supported_llms = ['gpt-3.5-turbo-16k', 'gpt-4', 'gpt-4-32k', 
+                      'api2d-gpt-3.5-turbo-16k', 'api2d-gpt-4']
     llm_kwargs['api_key'] = select_api_key(llm_kwargs['api_key'], llm_kwargs['llm_model'])
     if llm_kwargs['llm_model'] not in supported_llms:
         chatbot.append([f"处理任务: {txt}", f"当前插件只支持{str(supported_llms)}, 当前模型{llm_kwargs['llm_model']}."])

crazy_functions/语音助手.py

@@ -1,39 +1,14 @@
 from toolbox import update_ui
 from toolbox import CatchException, get_conf, markdown_convertion
 from crazy_functions.crazy_utils import input_clipping
+from crazy_functions.agent_fns.watchdog import WatchDog
 from request_llm.bridge_all import predict_no_ui_long_connection
 import threading, time
 import numpy as np
 from .live_audio.aliyunASR import AliyunASR
 import json
 import re
 
-class WatchDog():
-    def __init__(self, timeout, bark_fn, interval=3, msg="") -> None:
-        self.last_feed = None
-        self.timeout = timeout
-        self.bark_fn = bark_fn
-        self.interval = interval
-        self.msg = msg
-        self.kill_dog = False
-
-    def watch(self):
-        while True:
-            if self.kill_dog: break
-            if time.time() - self.last_feed > self.timeout:
-                if len(self.msg) > 0: print(self.msg)
-                self.bark_fn()
-                break
-            time.sleep(self.interval)
-
-    def begin_watch(self):
-        self.last_feed = time.time()
-        th = threading.Thread(target=self.watch)
-        th.daemon = True
-        th.start()
-
-    def feed(self):
-        self.last_feed = time.time()
 
 def chatbot2history(chatbot):
     history = []

docs/use_azure.md

@@ -1,3 +1,42 @@
+# 微软Azure云接入指南
+
+## 方法一（旧方法，只能接入一个Azure模型）
+
+- 通过以下教程，获取AZURE_ENDPOINT，AZURE_API_KEY，AZURE_ENGINE，直接修改 config 配置即可。配置的修改方法见本项目wiki。
+
+## 方法二（新方法，接入多个Azure模型，并支持动态切换）
+
+- 在方法一的基础上，注册并获取多组 AZURE_ENDPOINT，AZURE_API_KEY，AZURE_ENGINE
+- 修改config中的AZURE_CFG_ARRAY配置项，按照格式填入多个Azure模型的配置，如下所示：
+
+```
+AZURE_CFG_ARRAY = {
+    "azure-gpt-3.5": # 第一个模型，azure模型必须以"azure-"开头
+    {
+        "AZURE_ENDPOINT": "https://你亲手写的api名称.openai.azure.com/",
+        "AZURE_API_KEY": "cccccccccccccccccccccccccccccccc",
+        "AZURE_ENGINE": "填入你亲手写的部署名1",
+        "AZURE_MODEL_MAX_TOKEN": 4096,
+    },
+    "azure-gpt-4":  # 第二个模型，azure模型必须以"azure-"开头
+    {
+        "AZURE_ENDPOINT": "https://你亲手写的api名称.openai.azure.com/",
+        "AZURE_API_KEY": "dddddddddddddddddddddddddddddddd",
+        "AZURE_ENGINE": "填入你亲手写的部署名2",
+        "AZURE_MODEL_MAX_TOKEN": 8192,
+    },
+    "azure-gpt-3.5-16k":  # 第三个模型，azure模型必须以"azure-"开头
+    {
+        "AZURE_ENDPOINT": "https://你亲手写的api名称.openai.azure.com/",
+        "AZURE_API_KEY": "eeeeeeeeeeeeeeeeeeeeeeeeeeeeeeee",
+        "AZURE_ENGINE": "填入你亲手写的部署名3",
+        "AZURE_MODEL_MAX_TOKEN": 16384,
+    },
+}
+```
+
+
+
 # 通过微软Azure云服务申请 Openai API
 
 由于Openai和微软的关系，现在是可以通过微软的Azure云计算服务直接访问openai的api，免去了注册和网络的问题。

multi_language.py

@@ -13,6 +13,7 @@
 
         4. Run `python multi_language.py`. 
             Note: You need to run it multiple times to increase translation coverage because GPT makes mistakes sometimes.
+           (You can also run `CACHE_ONLY=True python multi_language.py` to use cached translation mapping)
 
         5. Find the translated program in `multi-language\English\*`
    
@@ -35,6 +36,8 @@
 import time
 from toolbox import get_conf
 
+CACHE_ONLY = os.environ.get('CACHE_ONLY', False)
+
 CACHE_FOLDER, = get_conf('PATH_LOGGING')
 
 blacklist = ['multi-language', CACHE_FOLDER, '.git', 'private_upload', 'multi_language.py', 'build', '.github', '.vscode', '__pycache__', 'venv']
@@ -336,7 +339,10 @@ def extract_chinese_characters_from_directory(directory_path):
         if d not in cached_translation_keys: 
             need_translate.append(d)
 
-    need_translate_mapping = trans(need_translate, language=LANG_STD, special=True)
+    if CACHE_ONLY:
+        need_translate_mapping = {}
+    else:
+        need_translate_mapping = trans(need_translate, language=LANG_STD, special=True)
     map_to_json(need_translate_mapping, language=LANG_STD)
     cached_translation = read_map_from_json(language=LANG_STD)
     cached_translation = dict(sorted(cached_translation.items(), key=lambda x: -len(x[0])))
@@ -476,8 +482,10 @@ def get_strings(node):
         if d not in cached_translation_keys: 
             need_translate.append(d)
 
-
-    up = trans_json(need_translate, language=LANG, special=False)
+    if CACHE_ONLY:
+        up = {}
+    else:
+        up = trans_json(need_translate, language=LANG, special=False)
     map_to_json(up, language=LANG)
     cached_translation = read_map_from_json(language=LANG)
     LANG_STD = 'std'