Merge pull request #584 from PiDanShouRouZhouXD/sakura_llm

Fix: Remove conflicting SakuraTranslator from translators
zyddnys · Feb 27, 2024 · 230ba8b · 230ba8b
2 parents f5dab5b + 3caf51f
commit 230ba8b
Show file tree

Hide file tree

Showing 5 changed files with 47 additions and 106 deletions.
diff --git a/README.md b/README.md
@@ -13,33 +13,41 @@
 
 Some manga/images will never be translated, therefore this project is born.
 
-- [Preview](#samples)
-- [Demo](#online-demo)
-- [Disclaimer](#disclaimer)
-- [Getting Started](#installation)
-    - [Installation](#installation)
-        - [Venv](#pipvenv)
-        - [Poetry](#poetry)
-        - [Extra Windows Info](#additional-instructions-for-windows)
-        - [Docker](#docker)
-    - [Usage](#usage)
-        - [Batch mode](#batch-mode-default)
-        - [Demo mode](#demo-mode)
-        - [Web mode](#web-mode)
-        - [Api mode](#api-mode)
-    - [Related Projects](#related-projects)
-- [Docs](#docs)
+- [Image/Manga Translator](#imagemanga-translator)
+  - [Samples](#samples)
+  - [Online Demo](#online-demo)
+  - [Disclaimer](#disclaimer)
+  - [Installation](#installation)
+    - [Pip/venv](#pipvenv)
+    - [Poetry](#poetry)
+      - [Additional instructions for **Windows**](#additional-instructions-for-windows)
+    - [Docker](#docker)
+      - [Hosting the web server](#hosting-the-web-server)
+      - [Using as CLI](#using-as-cli)
+      - [Setting Translation Secrets](#setting-translation-secrets)
+      - [Using with Nvidia GPU](#using-with-nvidia-gpu)
+      - [Building locally](#building-locally)
+  - [Usage](#usage)
+    - [Batch mode (default)](#batch-mode-default)
+    - [Demo mode](#demo-mode)
+    - [Web Mode](#web-mode)
+    - [Api Mode](#api-mode)
+  - [Related Projects](#related-projects)
+  - [Docs](#docs)
     - [Recommended Modules](#recommended-modules)
-    - [Args](#options)
-    - [Languages](#language-code-reference)
-    - [Translators](#translators-reference)
-    - [GPT config](#gpt-config-reference)
-    - [Gimp](#using-gimp-for-rendering)
-    - [Api Docs](#api-documentation)
-        - [v1](#api-documentation)
-        - [v2](#api-documentation)
-- [Roadmap](#next-steps)
-- [Support Us](#support-us)
+      - [Tips to improve translation quality](#tips-to-improve-translation-quality)
+    - [Options](#options)
+    - [Language Code Reference](#language-code-reference)
+    - [Translators Reference](#translators-reference)
+    - [GPT Config Reference](#gpt-config-reference)
+    - [Using Gimp for rendering](#using-gimp-for-rendering)
+    - [Api Documentation](#api-documentation)
+      - [Synchronous mode](#synchronous-mode)
+      - [Asynchronous mode](#asynchronous-mode)
+      - [Manual translation](#manual-translation)
+  - [Next steps](#next-steps)
+  - [Support Us](#support-us)
+    - [Thanks To All Our Contributors :](#thanks-to-all-our-contributors-)
 
 ## Samples
 
@@ -521,6 +529,7 @@ IND: Indonesian
 | gpt3.5     | ✔️      |         | Implements gpt-3.5-turbo. Requires `OPENAI_API_KEY`    |
 | gpt4       | ✔️      |         | Implements gpt-4. Requires `OPENAI_API_KEY`            |
 | papago     |         |         |                                                        |
+| sakura     |         |         |Requires `SAKURA_API_BASE`                               |
 | offline    |         | ✔️      | Chooses most suitable offline translator for language  |
 | sugoi      |         | ✔️      | Sugoi V4.0 Models                                      |
 | m2m100     |         | ✔️      | Supports every language                                |

diff --git a/README_CN.md b/README_CN.md
@@ -66,6 +66,7 @@ $ pip install git+https://github.com/kodalli/pydensecrf.git
 | gpt3.5         | ✔️      |         | Implements gpt-3.5-turbo. Requires `OPENAI_API_KEY`   |
 | gpt4           | ✔️      |         | Implements gpt-4. Requires `OPENAI_API_KEY`           |
 | papago         |         |         |                                                       |
+| sakura         |         |         |需要`SAKURA_API_BASE`|
 | offline        |         | ✔️      |  自动选择可用的离线模型，只是选择器                                                  |
 | sugoi          |         | ✔️      |  只能翻译英文                                                    |
 | m2m100         |         | ✔️      |  可以翻译所有语言                                                     |

diff --git a/manga_translator/translators/__init__.py b/manga_translator/translators/__init__.py
@@ -7,7 +7,7 @@
 from .deepl import DeeplTranslator
 from .papago import PapagoTranslator
 from .caiyun import CaiyunTranslator
-from .chatgpt import GPT3Translator, GPT35TurboTranslator, GPT4Translator, SakuraTranslator
+from .chatgpt import GPT3Translator, GPT35TurboTranslator, GPT4Translator
 from .nllb import NLLBTranslator, NLLBBigTranslator
 from .sugoi import JparacrawlTranslator, JparacrawlBigTranslator, SugoiTranslator
 from .m2m100 import M2M100Translator, M2M100BigTranslator
@@ -25,7 +25,6 @@
     'jparacrawl_big': JparacrawlBigTranslator,
     'm2m100': M2M100Translator,
     'm2m100_big': M2M100BigTranslator,
-    'sakura': SakuraTranslator,
 }
 
 TRANSLATORS = {

diff --git a/manga_translator/translators/chatgpt.py b/manga_translator/translators/chatgpt.py
@@ -9,8 +9,7 @@
 from typing import List, Dict
 
 from .common import CommonTranslator, MissingAPIKeyException
-from .keys import OPENAI_API_KEY, OPENAI_HTTP_PROXY, OPENAI_API_BASE, SAKURA_API_BASE, SAKURA_API_KEY
-
+from .keys import OPENAI_API_KEY, OPENAI_HTTP_PROXY, OPENAI_API_BASE
 CONFIG = None
 
 class GPT3Translator(CommonTranslator):
@@ -320,77 +319,3 @@ async def _request_translation(self, to_lang: str, prompt: str) -> str:
 
         # If no response with text is found, return the first response's content (which may be empty)
         return response.choices[0].message.content
-
-
-class SakuraTranslator(GPT3Translator):
-    _CONFIG_KEY = 'sakura'
-    _MAX_REQUESTS_PER_MINUTE = 200
-    _RETRY_ATTEMPTS = 5
-    _MAX_TOKENS = 8192
-    _CHAT_SYSTEM_TEMPLATE = (
-        '你是一个轻小说翻译模型，可以流畅通顺地以日本轻小说的风格将日文翻译成简体中文，并联系上下文正确使用人称代词，不擅自添加原文中没有的代词。'
-    )
-    def __init__(self):
-        super().__init__(check_openai_key=False)
-
-    async def _request_translation(self, to_lang: str, prompt: str) -> str:
-        messages = [
-            {'role': 'system', 'content': self._CHAT_SYSTEM_TEMPLATE},
-            {'role': 'user', 'content': '将下面的日文文本翻译成中文：'+prompt},
-        ]
-
-        response = await openai.ChatCompletion.acreate(
-            model='gpt-4-0613',
-            messages=messages,
-            max_tokens=self._MAX_TOKENS // 2,
-            temperature=self.temperature,
-            top_p=self.top_p,
-            api_key=SAKURA_API_KEY,
-            api_base=SAKURA_API_BASE,
-        )
-
-        self.token_count += response.usage['total_tokens']
-        self.token_count_last = response.usage['total_tokens']
-        for choice in response.choices:
-            if 'text' in choice:
-                return choice.text
-
-        # If no response with text is found, return the first response's content (which may be empty)
-        return response.choices[0].message.content
-    async def _translate(self, from_lang: str, to_lang: str, queries: List[str]) -> List[str]:
-        translations = []
-        self.logger.debug(f'Temperature: {self.temperature}, TopP: {self.top_p}')
-
-        for query in queries:
-
-            ratelimit_attempt = 0
-            server_error_attempt = 0
-            timeout_attempt = 0
-            while True:
-                request_task = asyncio.create_task(self._request_translation(to_lang, query))
-                try:
-                    response = await request_task
-                    break
-                except openai.error.RateLimitError: # Server returned ratelimit response
-                    ratelimit_attempt += 1
-                    if ratelimit_attempt >= self._RATELIMIT_RETRY_ATTEMPTS:
-                        raise
-                    self.logger.warn(f'Restarting request due to ratelimiting by openai servers. Attempt: {ratelimit_attempt}')
-                    await asyncio.sleep(2)
-                except openai.error.APIError: # Server returned 500 error (probably server load)
-                    server_error_attempt += 1
-                    if server_error_attempt >= self._RETRY_ATTEMPTS:
-                        self.logger.error('Sakura encountered a server error, possibly due to high server load. Use a different translator or try again later.')
-                        raise
-                    self.logger.warn(f'Restarting request due to a server error. Attempt: {server_error_attempt}')
-                    await asyncio.sleep(1)
-
-            self.logger.debug('-- Sakura Response --\n' + response)
-
-            translations.extend([response])
-
-        self.logger.debug(translations)
-        if self.token_count_last:
-            self.logger.info(f'Used {self.token_count_last} tokens (Total: {self.token_count})')
-
-        return translations
diff --git a/manga_translator/translators/sakura.py b/manga_translator/translators/sakura.py
@@ -17,7 +17,7 @@
 class Sakura13BTranslator(CommonTranslator):
 
     _TIMEOUT = 999  # Seconds to wait for a response from the server before retrying
-    _RETRY_ATTEMPTS = 3  # Number of times to retry an errored request before giving up
+    _RETRY_ATTEMPTS = 1  # Number of times to retry an errored request before giving up
     _TIMEOUT_RETRY_ATTEMPTS = 3  # Number of times to retry a timed out request before giving up
     _RATELIMIT_RETRY_ATTEMPTS = 3  # Number of times to retry a ratelimited request before giving up
 
@@ -128,6 +128,10 @@ async def _translate(self, from_lang: str, to_lang: str, queries: List[str]) ->
         self.logger.debug(f'Queries: {queries}')
         text_prompt = '\n'.join(queries)
         self.logger.debug('-- Sakura Prompt --\n' + self._format_prompt_log(text_prompt) + '\n\n')
+        # 去除emoji
+        queries = [re.sub(r'[\U00010000-\U0010ffff]', '', query) for query in queries]
+        # 替换❤
+        queries = [re.sub(r'❤', '♥', query) for query in queries]
         # 给queries的每行加上「」
         queries = [f'「{query}」' for query in queries]
         response = await self._handle_translation_request(queries)
@@ -139,6 +143,9 @@ async def _translate(self, from_lang: str, to_lang: str, queries: List[str]) ->
         rep_flag = self.detect_and_remove_extra_repeats(response)[0]
         if rep_flag:
             for i in range(self._RETRY_ATTEMPTS):
+                if self.detect_and_remove_extra_repeats(queries)[0]:
+                    self.logger.warning('Queries have repeats.')
+                    break
                 self.logger.warning(f'Re-translated because of model degradation, {i} times.')
                 self._set_gpt_style("precise")
                 self.logger.debug(f'Temperature: {self.temperature}, TopP: {self.top_p}')
@@ -254,7 +261,7 @@ async def _request_translation(self, input_text_list) -> str:
             ],
             temperature=self.temperature,
             top_p=self.top_p,
-            max_tokens=512,
+            max_tokens=1024,
             frequency_penalty=self.frequency_penalty,
             seed=-1,
             extra_query=extra_query,