feat: new openai package

Signed-off-by: yihong0618 <zouzou0208@gmail.com>

README-CN.md

@@ -19,7 +19,7 @@ bilingual_book_maker 是一个 AI 翻译工具，使用 ChatGPT 帮助用户制
 - 使用 `--openai_key` 指定 OpenAI API key，如果有多个可以用英文逗号分隔(xxx,xxx,xxx)，可以减少接口调用次数限制带来的错误。  
    或者，指定环境变量 `BBM_OPENAI_API_KEY` 来略过这个选项。
 - 本地放了一个 `test_books/animal_farm.epub` 给大家测试
-- 默认用了 [GPT-3.5-turbo](https://openai.com/blog/introducing-chatgpt-and-whisper-apis) 模型，也就是 ChatGPT 正在使用的模型，用 `--model gpt3` 来使用 gpt3 模型
+- 默认用了 [GPT-3.5-turbo](https://openai.com/blog/introducing-chatgpt-and-whisper-apis) 模型，也就是 ChatGPT 正在使用的模型。
 - 可以使用 DeepL 封装的 api 进行翻译，需要付费，[DeepL Translator](https://rapidapi.com/splintPRO/api/dpl-translator) 来获得 token  `--model deepl --deepl_key ${deepl_key}`
 - 可以使用 DeepL free `--model deeplfree`
 - 可以使用 [Claude](https://console.anthropic.com/docs) 模型进行翻译 `--model claude --claude_key ${claude_key}`
@@ -60,9 +60,6 @@ python3 make_book.py --book_name test_books/animal_farm.epub --openai_key ${open
 # 指定环境变量来略过 --openai_key
 export OPENAI_API_KEY=${your_api_key}
 
-# 或使用 gpt3 模型
-python3 make_book.py --book_name test_books/animal_farm.epub --model gpt3 --language ja
-
 # Use the DeepL model with Japanese
 python3 make_book.py --book_name test_books/animal_farm.epub --model deepl --deepl_key ${deepl_key} --language ja
 

README.md

@@ -24,7 +24,7 @@ Find more info here for using liteLLM: https://github.com/BerriAI/litellm/blob/m
 - Use `--openai_key` option to specify OpenAI API key. If you have multiple keys, separate them by commas (xxx,xxx,xxx) to reduce errors caused by API call limits.
    Or, just set environment variable `BBM_OPENAI_API_KEY` instead.
 - A sample book, `test_books/animal_farm.epub`, is provided for testing purposes.
-- The default underlying model is [GPT-3.5-turbo](https://openai.com/blog/introducing-chatgpt-and-whisper-apis), which is used by ChatGPT currently. Use `--model gpt4` to change the underlying model to `GPT4` and use `--model gpt3` to change the model to `GPT3`.
+- The default underlying model is [GPT-3.5-turbo](https://openai.com/blog/introducing-chatgpt-and-whisper-apis), which is used by ChatGPT currently. Use `--model gpt4` to change the underlying model to `GPT4`.
    If using `GPT4`, you can add `--use_context` to add a context paragraph to each passage sent to the model for translation (see below)
 - support DeepL model [DeepL Translator](https://rapidapi.com/splintPRO/api/dpl-translator) need pay to get the token use `--model deepl --deepl_key ${deepl_key}`
 - support DeepL free model `--model deeplfree`
@@ -78,9 +78,6 @@ export OPENAI_API_KEY=${your_api_key}
 # Use the GPT-4 model with context to Japanese
 python3 make_book.py --book_name test_books/animal_farm.epub --model gpt4 --use_context --language ja
 
-# Use the GPT-3 model with Japanese
-python3 make_book.py --book_name test_books/animal_farm.epub --model gpt3 --language ja
-
 # Use the DeepL model with Japanese
 python3 make_book.py --book_name test_books/animal_farm.epub --model deepl --deepl_key ${deepl_key} --language ja
 

book_maker/cli.py

@@ -259,7 +259,7 @@ So you are close to reaching the limit. You have to choose your own value, there
         "--temperature",
         type=float,
         default=1.0,
-        help="temperature parameter for `gpt3`/`chatgptapi`/`gpt4`/`claude`",
+        help="temperature parameter for `chatgptapi`/`gpt4`/`claude`",
     )
 
     options = parser.parse_args()
@@ -276,7 +276,7 @@ So you are close to reaching the limit. You have to choose your own value, there
     translate_model = MODEL_DICT.get(options.model)
     assert translate_model is not None, "unsupported model"
     API_KEY = ""
-    if options.model in ["gpt3", "chatgptapi", "gpt4"]:
+    if options.model in ["chatgptapi", "gpt4"]:
         if OPENAI_API_KEY := (
             options.openai_key
             or env.get(
@@ -287,6 +287,7 @@ So you are close to reaching the limit. You have to choose your own value, there
             )  # suggest adding `BBM_` prefix for all the bilingual_book_maker ENVs.
         ):
             API_KEY = OPENAI_API_KEY
+            # patch
         else:
             raise Exception(
                 "OpenAI API key not provided, please google how to obtain it",
@@ -373,12 +374,16 @@ So you are close to reaching the limit. You have to choose your own value, there
     if options.deployment_id:
         # only work for ChatGPT api for now
         # later maybe support others
-        assert (
+        assert options.model in [
-            options.model == "chatgptapi"
+            "chatgptapi",
-        ), "only support chatgptapi for deployment_id"
+            "gpt4",
+        ], "only support chatgptapi for deployment_id"
         if not options.api_base:
             raise ValueError("`api_base` must be provided when using `deployment_id`")
         e.translate_model.set_deployment_id(options.deployment_id)
+    # TODO refactor, quick fix for gpt4 model
+    if options.model == "gpt4":
+        e.translate_model.set_gpt4_models("gpt4")
 
     e.make_bilingual_book()
 

book_maker/translator/__init__.py

@@ -3,19 +3,16 @@ from book_maker.translator.chatgptapi_translator import ChatGPTAPI
 from book_maker.translator.deepl_translator import DeepL
 from book_maker.translator.deepl_free_translator import DeepLFree
 from book_maker.translator.google_translator import Google
-from book_maker.translator.gpt3_translator import GPT3
-from book_maker.translator.gpt4_translator import GPT4
 from book_maker.translator.claude_translator import Claude
 from book_maker.translator.custom_api_translator import CustomAPI
 
 MODEL_DICT = {
     "chatgptapi": ChatGPTAPI,
-    "gpt3": GPT3,
     "google": Google,
     "caiyun": Caiyun,
     "deepl": DeepL,
     "deeplfree": DeepLFree,
-    "gpt4": GPT4,
+    "gpt4": ChatGPTAPI,
     "claude": Claude,
     "customapi": CustomAPI
     # add more here

book_maker/translator/chatgptapi_translator.py

@@ -2,9 +2,10 @@ import re
 import time
 from copy import copy
 from os import environ
-from rich import print
+from itertools import cycle
 
-import openai
+from openai import AzureOpenAI, OpenAI
+from rich import print
 
 from .base_translator import Base
 
@@ -13,6 +14,21 @@ PROMPT_ENV_MAP = {
     "system": "BBM_CHATGPTAPI_SYS_MSG",
 }
 
+GPT35_MODEL_LIST = [
+    "gpt-3.5-turbo",
+    "gpt-3.5-turbo-1106",
+    "gpt-3.5-turbo-16k",
+    "gpt-3.5-turbo-0613",
+    "gpt-3.5-turbo-16k-0613",
+    "gpt-3.5-turbo-0301",
+]
+GPT4_MODEL_LIST = [
+    "gpt-4-1106-preview",
+    "gpt-4",
+    "gpt-4-0613",
+    "gpt-4-0314",
+]
+
 
 class ChatGPTAPI(Base):
     DEFAULT_PROMPT = "Please help me to translate,`{text}` to {language}, please return only translated content not include the origin text"
@@ -29,9 +45,8 @@ class ChatGPTAPI(Base):
     ) -> None:
         super().__init__(key, language)
         self.key_len = len(key.split(","))
+        self.openai_client = OpenAI(api_key=key, base_url=api_base)
 
-        if api_base:
-            openai.api_base = api_base
         self.prompt_template = (
             prompt_template
             or environ.get(PROMPT_ENV_MAP["user"])
@@ -48,9 +63,15 @@ class ChatGPTAPI(Base):
         self.system_content = environ.get("OPENAI_API_SYS_MSG") or ""
         self.deployment_id = None
         self.temperature = temperature
+        # gpt3 all models for save the limit
+        self.model_list = cycle(GPT35_MODEL_LIST)
 
     def rotate_key(self):
-        openai.api_key = next(self.keys)
+        self.openai_client.api_key = next(self.keys)
+
+    def rotate_model(self):
+        # TODO
+        self.model = next(self.model_list)
 
     def create_chat_completion(self, text):
         content = self.prompt_template.format(
@@ -62,49 +83,25 @@ class ChatGPTAPI(Base):
             {"role": "user", "content": content},
         ]
 
-        if self.deployment_id:
+        completion = self.openai_client.chat.completions.create(
-            return openai.ChatCompletion.create(
+            model=self.model,
-                engine=self.deployment_id,
-                messages=messages,
-                temperature=self.temperature,
-            )
-
-        return openai.ChatCompletion.create(
-            model="gpt-3.5-turbo-16k",
             messages=messages,
             temperature=self.temperature,
         )
+        return completion
 
     def get_translation(self, text):
         self.rotate_key()
+        self.rotate_model()  # rotate all the model to aviod the limit
 
-        completion = {}
         try:
             completion = self.create_chat_completion(text)
-        except Exception:
+        except Exception as e:
-            if (
+            print(e)
-                "choices" not in completion
+            pass
-                or not isinstance(completion["choices"], list)
-                or len(completion["choices"]) == 0
-            ):
-                raise
-            if completion["choices"][0]["finish_reason"] != "length":
-                raise
 
-        # work well or exception finish by length limit
+        # TODO work well or exception finish by length limit
-        choice = completion["choices"][0]
+        t_text = completion.choices[0].message.content.encode("utf8").decode() or ""
-
-        t_text = choice.get("message").get("content", "").encode("utf8").decode()
-
-        if choice["finish_reason"] == "length":
-            with open("log/long_text.txt", "a") as f:
-                print(
-                    f"""==================================================
-The total token is too long and cannot be completely translated\n
-{text}
-""",
-                    file=f,
-                )
 
         return t_text
 
@@ -278,7 +275,7 @@ The total token is too long and cannot be completely translated\n
         result_list, retry_count = self.get_best_result_list(
             plist_len,
             new_str,
-            6,
+            6,  # WTF this magic number here?
             result_list,
         )
 
@@ -295,6 +292,13 @@ The total token is too long and cannot be completely translated\n
         return result_list
 
     def set_deployment_id(self, deployment_id):
-        openai.api_type = "azure"
-        openai.api_version = "2023-03-15-preview"
         self.deployment_id = deployment_id
+        self.openai_client = AzureOpenAI(
+            api_key=next(self.keys),
+            azure_endpoint=self.api_base,
+            api_version="2023-07-01-preview",
+            azure_deployment=self.deployment_id,
+        )
+
+    def set_gpt4_models(self, model="gpt4"):
+        self.model_list = cycle(GPT4_MODEL_LIST)

book_maker/translator/gpt3_translator.py

@@ -1,56 +0,0 @@
-import re
-import requests
-from rich import print
-
-from .base_translator import Base
-
-
-class GPT3(Base):
-    def __init__(
-        self,
-        key,
-        language,
-        api_base=None,
-        prompt_template=None,
-        temperature=1.0,
-        **kwargs,
-    ) -> None:
-        super().__init__(key, language)
-        self.api_url = (
-            f"{api_base}v1/completions"
-            if api_base
-            else "https://api.openai.com/v1/completions"
-        )
-        self.headers = {
-            "Content-Type": "application/json",
-        }
-        # TODO support more models here
-        self.data = {
-            "prompt": "",
-            "model": "text-davinci-003",
-            "max_tokens": 1024,
-            "temperature": temperature,
-            "top_p": 1,
-        }
-        self.session = requests.session()
-        self.language = language
-        self.prompt_template = (
-            prompt_template or "Please help me to translate, `{text}` to {language}"
-        )
-
-    def rotate_key(self):
-        self.headers["Authorization"] = f"Bearer {next(self.keys)}"
-
-    def translate(self, text):
-        print(text)
-        self.rotate_key()
-        self.data["prompt"] = self.prompt_template.format(
-            text=text,
-            language=self.language,
-        )
-        r = self.session.post(self.api_url, headers=self.headers, json=self.data)
-        if not r.ok:
-            return text
-        t_text = r.json().get("choices")[0].get("text", "").strip()
-        print("[bold green]" + re.sub("\n{3,}", "\n\n", t_text) + "[/bold green]")
-        return t_text

book_maker/translator/gpt4_translator.py

@@ -1,330 +0,0 @@
-import re
-import time
-from copy import copy
-from os import environ, linesep
-
-import openai
-from rich import print
-
-from .base_translator import Base
-
-PROMPT_ENV_MAP = {
-    "user": "BBM_CHATGPTAPI_USER_MSG_TEMPLATE",
-    "system": "BBM_CHATGPTAPI_SYS_MSG",
-}
-
-
-class GPT4(Base):
-    DEFAULT_PROMPT = "Please help me to translate,`{text}` to {language}, please return only translated content not include the origin text"
-
-    def __init__(
-        self,
-        key,
-        language,
-        api_base=None,
-        prompt_template=None,
-        prompt_sys_msg=None,
-        context_flag=False,
-        temperature=1.0,
-        **kwargs,
-    ) -> None:
-        super().__init__(key, language)
-        self.context_flag = context_flag
-        self.context = "<summary>The start of the story.</summary>"
-        self.key_len = len(key.split(","))
-
-        if api_base:
-            openai.api_base = api_base
-        self.prompt_template = (
-            prompt_template
-            or environ.get(PROMPT_ENV_MAP["user"])
-            or self.DEFAULT_PROMPT
-        )
-        self.prompt_sys_msg = (
-            prompt_sys_msg
-            or environ.get(
-                "OPENAI_API_SYS_MSG",
-            )  # XXX: for backward compatibility, deprecate soon
-            or environ.get(PROMPT_ENV_MAP["system"])
-            or ""
-        )
-        self.system_content = environ.get("OPENAI_API_SYS_MSG") or ""
-        self.deployment_id = None
-        self.temperature = temperature
-
-    def rotate_key(self):
-        openai.api_key = next(self.keys)
-
-    def create_chat_completion(self, text):
-        # content = self.prompt_template.format(
-        #     text=text, language=self.language, crlf="\n"
-        # )
-
-        content = f"{self.context if self.context_flag else ''} {self.prompt_template.format(text=text, language=self.language, crlf=linesep)}"
-
-        sys_content = self.system_content or self.prompt_sys_msg.format(crlf="\n")
-
-        context_sys_str = "For each passage given, you may be provided a summary of the story up until this point (wrapped in tags '<summary>' and '</summary>') for context within the query, to provide background context of the story up until this point. If it's provided, use the context summary to aid you in translation with deeper comprehension, and write a new summary above the returned translation, wrapped in '<summary>' HTML-like tags, including important details (if relevant) from the new passage, retaining the most important key details from the existing summary, and dropping out less important details. If the summary is blank, assume it is the start of the story and write a summary from scratch. Do not make the summary longer than a paragraph, and smaller details can be replaced based on the relative importance of new details. The summary should be formatted in straightforward, inornate text, briefly summarising the entire story (from the start, including information before the given passage, leading up to the given passage) to act as an instructional payload for a Large-Language AI Model to fully understand the context of the passage."
-
-        sys_content = f"{self.system_content or self.prompt_sys_msg.format(crlf=linesep)} {context_sys_str if self.context_flag else ''} "
-
-        messages = [
-            {"role": "system", "content": sys_content},
-            {"role": "user", "content": content},
-        ]
-
-        if self.deployment_id:
-            return openai.ChatCompletion.create(
-                engine=self.deployment_id,
-                messages=messages,
-                temperature=self.temperature,
-            )
-
-        return openai.ChatCompletion.create(
-            model="gpt-4-1106-preview",
-            messages=messages,
-            temperature=self.temperature,
-        )
-
-    def get_translation(self, text):
-        self.rotate_key()
-
-        completion = {}
-        try:
-            completion = self.create_chat_completion(text)
-        except Exception:
-            if (
-                "choices" not in completion
-                or not isinstance(completion["choices"], list)
-                or len(completion["choices"]) == 0
-            ):
-                raise
-            if completion["choices"][0]["finish_reason"] != "length":
-                raise
-
-        # work well or exception finish by length limit
-        choice = completion["choices"][0]
-
-        t_text = choice.get("message").get("content", "").encode("utf8").decode()
-
-        if choice["finish_reason"] == "length":
-            with open("log/long_text.txt", "a") as f:
-                print(
-                    f"""==================================================
-The total token is too long and cannot be completely translated\n
-{text}
-""",
-                    file=f,
-                )
-
-        return t_text
-
-    def translate(self, text, needprint=True):
-        # print("=================================================")
-        start_time = time.time()
-        # todo: Determine whether to print according to the cli option
-        if needprint:
-            print(re.sub("\n{3,}", "\n\n", text))
-
-        attempt_count = 0
-        max_attempts = 3
-        t_text = ""
-
-        while attempt_count < max_attempts:
-            try:
-                t_text = self.get_translation(text)
-
-                # Extract the text between <summary> and </summary> tags (including the tags), save the next context text, then delete it from the text.
-                context_match = re.search(
-                    r"(<summary>.*?</summary>)", t_text, re.DOTALL
-                )
-                if context_match:
-                    self.context = context_match.group(0)
-                    t_text = t_text.replace(self.context, "", 1)
-                else:
-                    pass
-                    # self.context = ""
-
-                break
-            except Exception as e:
-                # todo: better sleep time? why sleep alawys about key_len
-                # 1. openai server error or own network interruption, sleep for a fixed time
-                # 2. an apikey has no money or reach limit, don`t sleep, just replace it with another apikey
-                # 3. all apikey reach limit, then use current sleep
-                sleep_time = int(60 / self.key_len)
-                print(e, f"will sleep {sleep_time} seconds")
-                time.sleep(sleep_time)
-                attempt_count += 1
-                if attempt_count == max_attempts:
-                    print(f"Get {attempt_count} consecutive exceptions")
-                    raise
-
-        # todo: Determine whether to print according to the cli option
-        if needprint:
-            print("[bold green]" + re.sub("\n{3,}", "\n\n", t_text) + "[/bold green]")
-
-        time.time() - start_time
-        # print(f"translation time: {elapsed_time:.1f}s")
-
-        return t_text
-
-    def translate_and_split_lines(self, text):
-        result_str = self.translate(text, False)
-        lines = result_str.splitlines()
-        lines = [line.strip() for line in lines if line.strip() != ""]
-        return lines
-
-    def get_best_result_list(
-        self,
-        plist_len,
-        new_str,
-        sleep_dur,
-        result_list,
-        max_retries=15,
-    ):
-        if len(result_list) == plist_len:
-            return result_list, 0
-
-        best_result_list = result_list
-        retry_count = 0
-
-        while retry_count < max_retries and len(result_list) != plist_len:
-            print(
-                f"bug: {plist_len} -> {len(result_list)} : Number of paragraphs before and after translation",
-            )
-            print(f"sleep for {sleep_dur}s and retry {retry_count+1} ...")
-            time.sleep(sleep_dur)
-            retry_count += 1
-            result_list = self.translate_and_split_lines(new_str)
-            if (
-                len(result_list) == plist_len
-                or len(best_result_list) < len(result_list) <= plist_len
-                or (
-                    len(result_list) < len(best_result_list)
-                    and len(best_result_list) > plist_len
-                )
-            ):
-                best_result_list = result_list
-
-        return best_result_list, retry_count
-
-    def log_retry(self, state, retry_count, elapsed_time, log_path="log/buglog.txt"):
-        if retry_count == 0:
-            return
-        print(f"retry {state}")
-        with open(log_path, "a", encoding="utf-8") as f:
-            print(
-                f"retry {state}, count = {retry_count}, time = {elapsed_time:.1f}s",
-                file=f,
-            )
-
-    def log_translation_mismatch(
-        self,
-        plist_len,
-        result_list,
-        new_str,
-        sep,
-        log_path="log/buglog.txt",
-    ):
-        if len(result_list) == plist_len:
-            return
-        newlist = new_str.split(sep)
-        with open(log_path, "a", encoding="utf-8") as f:
-            print(f"problem size: {plist_len - len(result_list)}", file=f)
-            for i in range(len(newlist)):
-                print(newlist[i], file=f)
-                print(file=f)
-                if i < len(result_list):
-                    print("............................................", file=f)
-                    print(result_list[i], file=f)
-                    print(file=f)
-                print("=============================", file=f)
-
-        print(
-            f"bug: {plist_len} paragraphs of text translated into {len(result_list)} paragraphs",
-        )
-        print("continue")
-
-    def join_lines(self, text):
-        lines = text.splitlines()
-        new_lines = []
-        temp_line = []
-
-        # join
-        for line in lines:
-            if line.strip():
-                temp_line.append(line.strip())
-            else:
-                if temp_line:
-                    new_lines.append(" ".join(temp_line))
-                    temp_line = []
-                new_lines.append(line)
-
-        if temp_line:
-            new_lines.append(" ".join(temp_line))
-
-        text = "\n".join(new_lines)
-
-        # del ^M
-        text = text.replace("^M", "\r")
-        lines = text.splitlines()
-        filtered_lines = [line for line in lines if line.strip() != "\r"]
-        new_text = "\n".join(filtered_lines)
-
-        return new_text
-
-    def translate_list(self, plist, context_flag):
-        sep = "\n\n\n\n\n"
-        # new_str = sep.join([item.text for item in plist])
-
-        new_str = ""
-        i = 1
-        for p in plist:
-            temp_p = copy(p)
-            for sup in temp_p.find_all("sup"):
-                sup.extract()
-            new_str += f"({i}) {temp_p.get_text().strip()}{sep}"
-            i = i + 1
-
-        if new_str.endswith(sep):
-            new_str = new_str[: -len(sep)]
-
-        new_str = self.join_lines(new_str)
-
-        plist_len = len(plist)
-
-        print(f"plist len = {len(plist)}")
-
-        result_list = self.translate_and_split_lines(new_str)
-
-        start_time = time.time()
-
-        result_list, retry_count = self.get_best_result_list(
-            plist_len,
-            new_str,
-            6,
-            result_list,
-        )
-
-        end_time = time.time()
-
-        state = "fail" if len(result_list) != plist_len else "success"
-        log_path = "log/buglog.txt"
-
-        self.log_retry(state, retry_count, end_time - start_time, log_path)
-        self.log_translation_mismatch(plist_len, result_list, new_str, sep, log_path)
-
-        # del (num), num. sometime (num) will translated to num.
-        result_list = [re.sub(r"^(\(\d+\)|\d+\.|(\d+))\s*", "", s) for s in result_list]
-
-        # # Remove the context paragraph from the final output
-        # if self.context:
-        #     context_len = len(self.context)
-        #     result_list = [s[context_len:] if s.startswith(self.context) else s for s in result_list]
-
-        return result_list
-
-    def set_deployment_id(self, deployment_id):
-        openai.api_type = "azure"
-        openai.api_version = "2023-03-15-preview"
-        self.deployment_id = deployment_id

setup.py

@@ -3,7 +3,7 @@ from setuptools import find_packages, setup
 
 packages = [
     "bs4",
-    "openai==0.27.2",
+    "openai>=1.1.1",
     "litellm",
     "requests",
     "ebooklib",
@@ -17,7 +17,7 @@ packages = [
 setup(
     name="bbook_maker",
     description="The bilingual_book_maker is an AI translation tool that uses ChatGPT to assist users in creating multi-language versions of epub/txt files and books.",
-    version="0.5.1",
+    version="0.6.0",
     license="MIT",
     author="yihong0618",
     author_email="zouzou0208@gmail.com",