1.8 OpenAI 格式化输出

1. OpenAI SDK 更新到 1.42.0
2. OpenAI 接口支持格式化输出
3. OpenAI 切换为局部代理
4. 翻译长文本的截断段落长度可配置

setup.py

@@ -37,7 +37,7 @@
     # For a discussion on single-sourcing the version across setup.py and the
     # project code, see
     # https://packaging.python.org/en/latest/single_source_version.html
-    version='1.7',  # Required. eg. 1.2.3
+    version='1.8',  # Required. eg. 1.2.3
 
     # This is a one-line description or tagline of what your project does. This
     # corresponds to the "Summary" metadata field:
@@ -143,7 +143,7 @@
     #
     # For an analysis of "install_requires" vs pip's requirements files see:
     # https://packaging.python.org/en/latest/requirements.html
-    install_requires=['py-color-log>=1.0.4', 'requests>=2.22.0', 'openai==0.27.2', 'tencentcloud-sdk-python==3.0.681'],  # Optional
+    install_requires=['py-color-log>=1.0.4', 'requests>=2.22.0', 'openai==1.42.0', 'tencentcloud-sdk-python==3.0.681'],  # Optional
 
     # List additional groups of dependencies here (e.g. development
     # dependencies). Users will be able to install these using the "extras"

src/transgpt/_settings.py

@@ -5,3 +5,4 @@
 CHARSET = 'UTF-8'
 CRLF = "\n"
 DOUBLE_CRLF = "\n\n"
+LOCALHOST = '127.0.0.1'

src/transgpt/_trans_base.py

@@ -13,14 +13,16 @@ class BaseTranslation :
 
     __metaclass__ = ABCMeta # 定义为抽象类
 
-    def __init__(self, api_id, api_key) -> None :
+    def __init__(self, api_id, api_key, cut_len=500) -> None :
         """
         翻译文本
         :param api_id  : 接口 API_ID / APP_ID （ChatGPT 不需要）
         :param api_key : 接口 API_KEY / SECRET_KEY
+        :param cut_len : 自动切割长文本的每一段长度（取决于 API 接口限制每次翻译的字数）
         """
         self.api_id = api_id
         self.api_key = api_key
+        self.cut_len = cut_len
 
 
     def translate(self, content, from_lang='', to_lang='', savepath='', oncesave=False, args={}) -> str :
@@ -114,7 +116,7 @@ def _save_trans(self, content, savepath) :
 
     @abstractmethod
     def _len_limit(self) :
-        return 500  # API 接口限制每次翻译的字数
+        return self.cut_len
 
 
     @abstractmethod

src/transgpt/trans_baidu.py

@@ -18,8 +18,8 @@
 
 class BaiduTranslation(BaseTranslation) :
 
-    def __init__(self, api_id, api_key, api_url=BAIDU_API_URL) -> None :
-        BaseTranslation.__init__(self, api_id, api_key)
+    def __init__(self, api_id, api_key, api_url=BAIDU_API_URL, cut_len=2000) -> None :
+        BaseTranslation.__init__(self, api_id, api_key, cut_len)
         self.api_url = api_url
 
 
@@ -71,8 +71,3 @@ def _to_sign(self, data) :
             )).encode(encoding=CHARSET)
         ).hexdigest()
         return salt, sign
-
-
-    def _len_limit(self) :
-        return 2000
-

src/transgpt/trans_chatgpt.py

@@ -8,36 +8,44 @@
 
 import os
 import time
-import openai
+import httpx
+from openai import OpenAI
 from ._settings import *
 from ._trans_base import BaseTranslation
 
-HTTP_PROXY = "HTTP_PROXY"
-HTTPS_PROXY = "HTTPS_PROXY"
 
 # GPT 接口模型定义 https://platform.openai.com/docs/models/
-CHATGPT_35_TURBO = "gpt-3.5-turbo"      # 16K
-CHATGPT_4 = "gpt-4"                     # 8K
-CHATGPT_4_TRUBO = "gpt-4-turbo"         # 128K
-CHATGPT_4o = "gpt-4o"                   # 128K（静态模型，适合翻译）
-CHATGPT_4o_LATEST = "chatgpt-4o-latest" # 128K（动态模型，价格高）
-CHATGPT_4o_MINI = "gpt-4o-mini"         # 128K（翻译较弱）
+CHATGPT_35_TURBO = "gpt-3.5-turbo"          # 16K
+CHATGPT_4 = "gpt-4"                         # 8K
+CHATGPT_4_TRUBO = "gpt-4-turbo"             # 128K
+CHATGPT_4o = "gpt-4o"                       # 128K（静态模型，适合翻译）
+CHATGPT_4o_20240806 = "gpt-4o-2024-08-06"   # 128K（静态模型，100% 格式化输出）
+CHATGPT_4o_LATEST = "chatgpt-4o-latest"     # 128K（动态模型，价格高）
+CHATGPT_4o_MINI = "gpt-4o-mini"             # 128K（翻译较弱）
+
+HTTP_PROXY = "HTTP_PROXY"
+HTTPS_PROXY = "HTTPS_PROXY"
 
 ARG_ROLE = 'role'
 ARG_OPENAI_MODEL = 'openai_model'
 ARG_PROXY_IP = 'proxy_ip'
 ARG_PROXY_PORT = 'proxy_port'
+ARG_RSP_FORMAT = 'response_format'
 
 class ChatgptTranslation(BaseTranslation) :
 
     RETRY = 3
     RETRY_WAIT_SECONDS = 30
 
-    def __init__(self, openai_key, openai_model=CHATGPT_4o_MINI, proxy_ip='127.0.0.1', proxy_port=0) :
-        BaseTranslation.__init__(self, '', openai_key)
-        openai.api_key = openai_key
+    def __init__(self, openai_key, openai_model=CHATGPT_4o_MINI, proxy_ip='127.0.0.1', proxy_port=0, cut_len=500, response_format=None) :
+        BaseTranslation.__init__(self, '', openai_key, cut_len)
+        proxy = f"http://{proxy_ip}:{proxy_port}" if proxy_port > 0 else ""
+        proxies = { "http://": proxy, "https://": proxy, }
+        http_cli = httpx.Client(proxies=proxies) if proxy else None
+
+        self.ai_cli = OpenAI(api_key=openai_key, http_client=http_cli)
+        self.response_format = response_format
         self.model = openai_model or CHATGPT_4o_MINI
-        self.proxy = f"http://{proxy_ip}:{proxy_port}" if proxy_port > 0 else ""
 
 
     def _translate(self, segment, from_lang='英文', to_lang='中文', args={}) :
@@ -56,49 +64,27 @@ def _translate(self, segment, from_lang='英文', to_lang='中文', args={}) :
 
 
     def _ask_gpt(self, role_setting, segment) :
-        self._enable_proxy()
         msg = [
             role_setting, 
             { "role": "user", "content": segment }
         ]
-        rsp = self._wait_for_ask(msg)
-        rst = rsp.get("choices")[0]["message"]["content"]
-        self._disable_proxy()
-        return rst
+        answer = self._wait_for_ask(msg)
+        return answer
 
 
     def _wait_for_ask(self, question) :
-        rsp = ""
+        answer = ""
         for i in range(self.RETRY) :
             try :
-                rsp = openai.ChatCompletion.create(
+                rsp = self.ai_cli.chat.completions.create(
                     model=self.model,
-                    messages=question
+                    messages=question, 
+                    response_format=self.response_format
                 )
+                answer = rsp.choices[0].message.content
                 break
 
-            except Exception as e:
-                rsp = {
-                    'choices': [{
-                        'message': {
-                            'content': f"[ERROR] ChatGPT No Response: {e}"
-                        }
-                    }]
-                }
+            except Exception as e :
+                answer = f"[ERROR] ChatGPT No Response: {e}"
                 time.sleep(self.RETRY_WAIT_SECONDS)
-        return rsp
-
-
-    def _enable_proxy(self) :
-        os.environ[HTTP_PROXY] = self.proxy
-        os.environ[HTTPS_PROXY] = self.proxy
-
-
-    def _disable_proxy(self) :
-        os.environ[HTTP_PROXY] = ""
-        os.environ[HTTPS_PROXY] = ""
-
-
-    def _len_limit(self) :
-        return 500
-
+        return answer

src/transgpt/trans_tencent.py

@@ -17,8 +17,8 @@
 
 class TencentTranslation(BaseTranslation) :
 
-    def __init__(self, api_id, api_key, region=GZ_REGION) :
-        BaseTranslation.__init__(self, api_id, api_key)
+    def __init__(self, api_id, api_key, region=GZ_REGION, cut_len=2000) :
+        BaseTranslation.__init__(self, api_id, api_key, cut_len)
         cred = Credential(api_id, api_key)
         self.client = TmtClient(cred, region)
 
@@ -43,8 +43,4 @@ def _translate(self, segment, from_lang='en', to_lang='zh', args={}) :
         rsp = self.client.TextTranslate(req)
         return rsp.TargetText
 
-
-    def _len_limit(self) :
-        return 2000
-
 

src/transgpt/translate.py

@@ -20,6 +20,7 @@
 def trans(content, api_id, api_key, 
             from_lang='', to_lang='', 
             platform=TENCENT, savepath='', oncesave=False, 
+            cut_len=500, 
             args={}
     ) -> str:
     """
@@ -32,6 +33,7 @@ def trans(content, api_id, api_key,
     :param platform     : 翻译平台，目前支持： chatgpt, baidu, tencent（默认）
     :param savepath     : 翻译文本保存的位置，若为空则不保存（可通过返回值获取）
     :param oncesave     : 是否一次性保存翻译文本（对于超长文本，内部会进行分段翻译，为了避免网络异常导致已翻译文本丢失，此项默认关闭）
+    :param cut_len      : 自动切割长文本的每一段长度（取决于 API 接口限制每次翻译的字数）
     :param args         : hash 其他参数表
                             api_url         : 仅百度翻译有用：接口 API 路径
                             region          : 仅腾讯翻译有用：翻译服务器所在的区域
@@ -46,17 +48,18 @@ def trans(content, api_id, api_key,
     log.info(f"正在使用 [{platform}] 翻译文本 ...")
     if platform == BAIDU :
         api_url = args.get(ARG_API_URL) or BAIDU_API_URL
-        client = BaiduTranslation(api_id, api_key, api_url)
+        client = BaiduTranslation(api_id, api_key, api_url, cut_len)
 
     elif platform == CHATGPT :
-        openai_model = args.get(ARG_OPENAI_MODEL) or CHATGPT_35_TURBO
-        proxy_ip = args.get(ARG_PROXY_IP) or '127.0.0.1'
+        openai_model = args.get(ARG_OPENAI_MODEL) or CHATGPT_4o_MINI
+        proxy_ip = args.get(ARG_PROXY_IP) or LOCALHOST
         proxy_port = args.get(ARG_PROXY_PORT) or 0
-        client = ChatgptTranslation(api_key, openai_model, proxy_ip, proxy_port)
+        response_format = args.get(ARG_RSP_FORMAT)
+        client = ChatgptTranslation(api_key, openai_model, proxy_ip, proxy_port, cut_len, response_format)
 
     else :
         region = args.get(ARG_REGION) or GZ_REGION
-        client = TencentTranslation(api_id, api_key, region)
+        client = TencentTranslation(api_id, api_key, region, cut_len)
 
     trans_content = client.translate(content, from_lang, to_lang, savepath, oncesave, args)
     log.info(f"文本翻译完成")