epub-translator 0.1.5__tar.gz → 0.1.6__tar.gz

{epub_translator-0.1.5 → epub_translator-0.1.6}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: epub-translator
-Version: 0.1.5
+Version: 0.1.6
 Summary: Translate the epub book using LLM. The translated book will retain the original text and list the translated text side by side with the original text.
 License: MIT
 Keywords: epub,llm,translation,translator
@@ -156,7 +156,8 @@ translate(
     submit: SubmitKind,                # How to insert translations (REPLACE, APPEND_TEXT, or APPEND_BLOCK)
     user_prompt: str | None = None,    # Custom translation instructions
     max_retries: int = 5,              # Maximum retries for failed translations
-    max_group_tokens: int = 1200,      # Maximum tokens per translation group
+    max_group_tokens: int = 2600,      # Maximum tokens per translation group
+    concurrency: int = 1,              # Number of concurrent translation tasks (default: 1)
     llm: LLM | None = None,            # Single LLM instance for both translation and filling
     translation_llm: LLM | None = None,  # LLM instance for translation (overrides llm)
     fill_llm: LLM | None = None,       # LLM instance for XML filling (overrides llm)
@@ -394,6 +395,32 @@ llm = LLM(
 )
 ```
 
+### Concurrent Translation
+
+Speed up translation by processing multiple text segments concurrently. Use the `concurrency` parameter to control how many translation tasks run in parallel:
+
+```python
+translate(
+    source_path="source.epub",
+    target_path="translated.epub",
+    target_language="English",
+    submit=SubmitKind.APPEND_BLOCK,
+    llm=llm,
+    concurrency=4,  # Process 4 segments concurrently
+)
+```
+
+**Performance Tips:**
+
+- Start with `concurrency=4` and adjust based on your API rate limits and system resources
+- Higher concurrency values can significantly reduce translation time for large books
+- The translation order is preserved regardless of concurrency settings
+- Monitor your API provider's rate limits to avoid throttling
+
+**Thread Safety:**
+
+When using `concurrency > 1`, ensure that any custom callback functions (`on_progress`, `on_fill_failed`) are thread-safe. Built-in callbacks are thread-safe by default.
+
 ## Related Projects
 
 ### PDF Craft

{epub_translator-0.1.5 → epub_translator-0.1.6}/README.md

@@ -123,7 +123,8 @@ translate(
     submit: SubmitKind,                # How to insert translations (REPLACE, APPEND_TEXT, or APPEND_BLOCK)
     user_prompt: str | None = None,    # Custom translation instructions
     max_retries: int = 5,              # Maximum retries for failed translations
-    max_group_tokens: int = 1200,      # Maximum tokens per translation group
+    max_group_tokens: int = 2600,      # Maximum tokens per translation group
+    concurrency: int = 1,              # Number of concurrent translation tasks (default: 1)
     llm: LLM | None = None,            # Single LLM instance for both translation and filling
     translation_llm: LLM | None = None,  # LLM instance for translation (overrides llm)
     fill_llm: LLM | None = None,       # LLM instance for XML filling (overrides llm)
@@ -361,6 +362,32 @@ llm = LLM(
 )
 ```
 
+### Concurrent Translation
+
+Speed up translation by processing multiple text segments concurrently. Use the `concurrency` parameter to control how many translation tasks run in parallel:
+
+```python
+translate(
+    source_path="source.epub",
+    target_path="translated.epub",
+    target_language="English",
+    submit=SubmitKind.APPEND_BLOCK,
+    llm=llm,
+    concurrency=4,  # Process 4 segments concurrently
+)
+```
+
+**Performance Tips:**
+
+- Start with `concurrency=4` and adjust based on your API rate limits and system resources
+- Higher concurrency values can significantly reduce translation time for large books
+- The translation order is preserved regardless of concurrency settings
+- Monitor your API provider's rate limits to avoid throttling
+
+**Thread Safety:**
+
+When using `concurrency > 1`, ensure that any custom callback functions (`on_progress`, `on_fill_failed`) are thread-safe. Built-in callbacks are thread-safe by default.
+
 ## Related Projects
 
 ### PDF Craft

{epub_translator-0.1.5 → epub_translator-0.1.6}/epub_translator/__init__.py

@@ -1,6 +1,5 @@
-from . import language
 from .llm import LLM
-from .translator import FillFailedEvent, translate
+from .translation import FillFailedEvent, language, translate
 from .xml_translator import SubmitKind
 
 __all__ = [

{epub_translator-0.1.5 → epub_translator-0.1.6}/epub_translator/epub/__init__.py

@@ -1,4 +1,4 @@
 from .metadata import read_metadata, write_metadata
 from .spines import search_spine_paths
-from .toc import read_toc, write_toc
+from .toc import Toc, read_toc, write_toc
 from .zip import Zip

{epub_translator-0.1.5 → epub_translator-0.1.6}/epub_translator/llm/context.py

@@ -1,5 +1,6 @@
 import hashlib
 import json
+import threading
 import uuid
 from pathlib import Path
 from typing import Self
@@ -8,6 +9,9 @@ from .executor import LLMExecutor
 from .increasable import Increasable, Increaser
 from .types import Message, MessageRole
 
+# Global lock for cache file commit operations
+_CACHE_COMMIT_LOCK = threading.Lock()
+
 
 class LLMContext:
     def __init__(
@@ -101,7 +105,12 @@ class LLMContext:
                 # Remove the .[context-id].txt suffix to get permanent name
                 permanent_name = temp_file.name.rsplit(".", 2)[0] + ".txt"
                 permanent_file = temp_file.parent / permanent_name
-                temp_file.rename(permanent_file)
+
+                with _CACHE_COMMIT_LOCK:  # 多线程下的线程安全
+                    if permanent_file.exists():
+                        temp_file.unlink()
+                    else:
+                        temp_file.rename(permanent_file)
 
     def _rollback(self) -> None:
         for temp_file in self._temp_files:

{epub_translator-0.1.5 → epub_translator-0.1.6}/epub_translator/llm/core.py

@@ -1,4 +1,5 @@
 import datetime
+import threading
 from collections.abc import Generator
 from importlib.resources import files
 from logging import DEBUG, FileHandler, Formatter, Logger, getLogger
@@ -14,6 +15,11 @@ from .executor import LLMExecutor
 from .increasable import Increasable
 from .types import Message
 
+# Global state for logger filename generation
+_LOGGER_LOCK = threading.Lock()
+_LAST_TIMESTAMP: str | None = None
+_LOGGER_SUFFIX_ID: int = 1
+
 
 class LLM:
     def __init__(
@@ -95,13 +101,34 @@ class LLM:
         return dir_path.resolve()
 
     def _create_logger(self) -> Logger | None:
+        # pylint: disable=global-statement
+        global _LAST_TIMESTAMP, _LOGGER_SUFFIX_ID
+
         if self._logger_save_path is None:
             return None
 
         now = datetime.datetime.now(datetime.UTC)
-        timestamp = now.strftime("%Y-%m-%d %H-%M-%S %f")
-        file_path = self._logger_save_path / f"request {timestamp}.log"
-        logger = getLogger(f"LLM Request {timestamp}")
+        # Use second-level precision for collision detection
+        timestamp_key = now.strftime("%Y-%m-%d %H-%M-%S")
+
+        with _LOGGER_LOCK:
+            if _LAST_TIMESTAMP == timestamp_key:
+                _LOGGER_SUFFIX_ID += 1
+                suffix_id = _LOGGER_SUFFIX_ID
+            else:
+                _LAST_TIMESTAMP = timestamp_key
+                _LOGGER_SUFFIX_ID = 1
+                suffix_id = 1
+
+        if suffix_id == 1:
+            file_name = f"request {timestamp_key}.log"
+            logger_name = f"LLM Request {timestamp_key}"
+        else:
+            file_name = f"request {timestamp_key}_{suffix_id}.log"
+            logger_name = f"LLM Request {timestamp_key}_{suffix_id}"
+
+        file_path = self._logger_save_path / file_name
+        logger = getLogger(logger_name)
         logger.setLevel(DEBUG)
         handler = FileHandler(file_path, encoding="utf-8")
         handler.setLevel(DEBUG)

{epub_translator-0.1.5 → epub_translator-0.1.6}/epub_translator/segment/inline_segment.py

@@ -47,6 +47,7 @@ def search_inline_segments(text_segments: Iterable[TextSegment]) -> Generator["I
                 inline_segment = _pop_stack_data(stack_data)
                 stack_data = None
                 if inline_segment:
+                    inline_segment.id = 0
                     yield inline_segment
 
         if stack_data is None:
@@ -73,6 +74,7 @@ def search_inline_segments(text_segments: Iterable[TextSegment]) -> Generator["I
     if stack_data is not None:
         inline_segment = _pop_stack_data(stack_data)
         if inline_segment:
+            inline_segment.id = 0
             yield inline_segment
 
 
@@ -115,7 +117,7 @@ class InlineSegment:
         self._child_tag2ids: dict[str, list[int]] = {}
         self._child_tag2count: dict[str, int] = {}
 
-        next_temp_id: int = 0
+        next_temp_id: int = 1
         terms = nest((child.parent.tag, child) for child in children if isinstance(child, InlineSegment))
 
         for tag, child_terms in terms.items():
@@ -162,6 +164,14 @@ class InlineSegment:
             elif isinstance(child, InlineSegment):
                 yield from child
 
+    def clone(self) -> "InlineSegment":
+        cloned_segment = InlineSegment(
+            depth=len(self._parent_stack),
+            children=[child.clone() for child in self._children],
+        )
+        cloned_segment.id = self.id
+        return cloned_segment
+
     def recreate_ids(self, id_generator: IDGenerator) -> None:
         self._child_tag2count.clear()
         self._child_tag2ids.clear()

{epub_translator-0.1.5 → epub_translator-0.1.6}/epub_translator/segment/text_segment.py

@@ -33,10 +33,6 @@ class TextSegment:
     def block_parent(self) -> Element:
         return self.parent_stack[self.block_depth - 1]
 
-    @property
-    def xml_text(self) -> str:
-        return "".join(_expand_xml_texts(self))
-
     def strip_block_parents(self) -> Self:
         self.parent_stack = self.parent_stack[self.block_depth - 1 :]
         self.block_depth = 1

epub_translator-0.1.6/epub_translator/translation/__init__.py

@@ -0,0 +1,2 @@
+from . import language
+from .translator import FillFailedEvent, translate

{epub_translator-0.1.5/epub_translator → epub_translator-0.1.6/epub_translator/translation}/epub_transcode.py

@@ -6,8 +6,8 @@ EPUB 数据结构与 XML 的编码/解码转换
 
 from xml.etree.ElementTree import Element
 
-from .epub.metadata import MetadataField
-from .epub.toc import Toc
+from ..epub import Toc
+from ..epub.metadata import MetadataField
 
 
 def encode_toc(toc: Toc) -> Element:

{epub_translator-0.1.5/epub_translator → epub_translator-0.1.6/epub_translator/translation}/punctuation.py

@@ -1,6 +1,6 @@
 from xml.etree.ElementTree import Element
 
-from .xml import iter_with_stack
+from ..xml import iter_with_stack
 
 _QUOTE_MAPPING = {
     # 法语引号

{epub_translator-0.1.5/epub_translator → epub_translator-0.1.6/epub_translator/translation}/translator.py

@@ -5,7 +5,7 @@ from importlib.metadata import version as get_package_version
 from os import PathLike
 from pathlib import Path
 
-from .epub import (
+from ..epub import (
     Zip,
     read_metadata,
     read_toc,
@@ -13,12 +13,12 @@ from .epub import (
     write_metadata,
     write_toc,
 )
+from ..llm import LLM
+from ..xml import XMLLikeNode, deduplicate_ids_in_element, find_first
+from ..xml_translator import FillFailedEvent, SubmitKind, TranslationTask, XMLTranslator
 from .epub_transcode import decode_metadata, decode_toc_list, encode_metadata, encode_toc_list
-from .llm import LLM
 from .punctuation import unwrap_french_quotes
-from .xml import XMLLikeNode, deduplicate_ids_in_element, find_first
 from .xml_interrupter import XMLInterrupter
-from .xml_translator import FillFailedEvent, SubmitKind, TranslationTask, XMLTranslator
 
 
 class _ElementType(Enum):
@@ -40,7 +40,8 @@ def translate(
     submit: SubmitKind,
     user_prompt: str | None = None,
     max_retries: int = 5,
-    max_group_tokens: int = 1200,
+    max_group_tokens: int = 2600,
+    concurrency: int = 1,
     llm: LLM | None = None,
     translation_llm: LLM | None = None,
     fill_llm: LLM | None = None,
@@ -62,7 +63,7 @@ def translate(
         ignore_translated_error=False,
         max_retries=max_retries,
         max_fill_displaying_errors=10,
-        max_group_tokens=max_group_tokens,
+        max_group_score=max_group_tokens,
         cache_seed_content=f"{_get_version()}:{target_language}",
     )
     with Zip(
@@ -92,6 +93,7 @@ def translate(
         current_progress = 0.0
 
         for translated_elem, context in translator.translate_elements(
+            concurrency=concurrency,
             interrupt_source_text_segments=interrupter.interrupt_source_text_segments,
             interrupt_translated_text_segments=interrupter.interrupt_translated_text_segments,
             interrupt_block_element=interrupter.interrupt_block_element,

{epub_translator-0.1.5/epub_translator → epub_translator-0.1.6/epub_translator/translation}/xml_interrupter.py

@@ -2,8 +2,8 @@ from collections.abc import Generator, Iterable
 from typing import cast
 from xml.etree.ElementTree import Element
 
-from .segment import TextSegment
-from .utils import ensure_list, normalize_whitespace
+from ..segment import TextSegment
+from ..utils import ensure_list, normalize_whitespace
 
 _ID_KEY = "__XML_INTERRUPTER_ID"
 _MATH_TAG = "math"

{epub_translator-0.1.5 → epub_translator-0.1.6}/epub_translator/xml/__init__.py

@@ -1,6 +1,6 @@
 from .const import *
 from .deduplication import *
-from .firendly import *
+from .friendly import *
 from .inline import *
 from .utils import *
 from .xml import *

epub_translator-0.1.6/epub_translator/xml_translator/concurrency.py

@@ -0,0 +1,52 @@
+from collections import deque
+from collections.abc import Callable, Iterable
+from concurrent.futures import Future, ThreadPoolExecutor
+from typing import TypeVar
+
+P = TypeVar("P")
+R = TypeVar("R")
+
+
+def run_concurrency(
+    parameters: Iterable[P],
+    execute: Callable[[P], R],
+    concurrency: int,
+) -> Iterable[R]:
+    assert concurrency >= 1, "the concurrency must be at least 1"
+    # Fast path: concurrency == 1, no thread overhead
+    if concurrency == 1:
+        for param in parameters:
+            yield execute(param)
+        return
+
+    executor = ThreadPoolExecutor(max_workers=concurrency)
+    did_shutdown = False
+    try:
+        futures: deque[Future[R]] = deque()
+        params_iter = iter(parameters)
+        for _ in range(concurrency):
+            try:
+                param = next(params_iter)
+                future = executor.submit(execute, param)
+                futures.append(future)
+            except StopIteration:
+                break
+
+        while futures:
+            future = futures.popleft()
+            yield future.result()
+            try:
+                param = next(params_iter)
+                new_future = executor.submit(execute, param)
+                futures.append(new_future)
+            except StopIteration:
+                pass
+
+    except KeyboardInterrupt:
+        executor.shutdown(wait=False, cancel_futures=True)
+        did_shutdown = True
+        raise
+
+    finally:
+        if not did_shutdown:
+            executor.shutdown(wait=True)

epub_translator-0.1.6/epub_translator/xml_translator/score.py

@@ -0,0 +1,164 @@
+from collections.abc import Generator
+from dataclasses import dataclass
+from enum import Enum, auto
+
+from tiktoken import Encoding
+
+from ..segment import InlineSegment, TextSegment
+from .common import DATA_ORIGIN_LEN_KEY
+
+_ID_WEIGHT = 80
+_ELLIPSIS = "..."
+
+
+@dataclass
+class ScoreSegment:
+    text_segment: TextSegment
+    left_parents: list[InlineSegment]
+    right_parents: list[InlineSegment]
+    text_tokens: list[int]
+    score: int
+
+
+def expand_to_score_segments(encoding: Encoding, inline_segment: InlineSegment) -> Generator[ScoreSegment, None, None]:
+    for i, score_segment in enumerate(_do_expand_inline_segment(inline_segment)):
+        xml_text = "".join(
+            _render_score_segment(
+                score_segment=score_segment,
+                is_first=(i == 0),
+            )
+        )
+        score_segment.text_tokens = encoding.encode(score_segment.text_segment.text)
+        score_segment.score = len(encoding.encode(xml_text)) + sum(
+            _ID_WEIGHT for parent in score_segment.left_parents if parent.id is not None
+        )
+        yield score_segment
+
+
+def truncate_score_segment(
+    encoding: Encoding,
+    score_segment: ScoreSegment,
+    remain_head: bool,
+    remain_score: int,
+):
+    fixed_score = score_segment.score - len(score_segment.text_tokens)
+    if remain_score <= fixed_score:
+        # 裁剪仅能减少 text 部分的 tokens 数。
+        # 而 XML 本身头尾占用的 tokens 数，以及 ID 占用加权分属于 fixed_score 部分，无法裁剪
+        # 当发现将文字删光后才能达标时，不如直接放弃整段内容
+        return None
+
+    remain_text_tokens_count = remain_score - fixed_score
+
+    # remain_text_tokens_count cannot be 0 here
+    if remain_head:
+        remain_text = encoding.decode(score_segment.text_tokens[:remain_text_tokens_count])
+    else:
+        remain_text = encoding.decode(score_segment.text_tokens[-remain_text_tokens_count:])
+
+    if not remain_text.strip():
+        return None
+
+    if remain_head:
+        remain_text = f"{remain_text} {_ELLIPSIS}"
+    else:
+        remain_text = f"{_ELLIPSIS} {remain_text}"
+
+    text_segment = score_segment.text_segment.clone()
+    text_segment.text = remain_text
+
+    return ScoreSegment(
+        text_segment=text_segment,
+        left_parents=score_segment.left_parents,
+        right_parents=score_segment.right_parents,
+        text_tokens=encoding.encode(remain_text),
+        score=remain_text_tokens_count + fixed_score,
+    )
+
+
+def _render_score_segment(score_segment: ScoreSegment, is_first: bool):
+    for i, parent in enumerate(score_segment.left_parents):
+        yield "<"
+        yield parent.parent.tag
+        if parent.id is not None:
+            yield ' id="99"'
+        if is_first and i == 0:
+            yield " "
+            yield DATA_ORIGIN_LEN_KEY
+            yield '="9999"'
+        yield ">"
+
+    yield score_segment.text_segment.text
+
+    for parent in reversed(score_segment.right_parents):
+        yield "</"
+        yield parent.parent.tag
+        yield ">"
+
+
+def _do_expand_inline_segment(inline_segment: InlineSegment):
+    text_segment: TextSegment | None = None
+    left_parents: list[InlineSegment] = []
+    right_parents: list[InlineSegment] = []
+
+    for item in _expand_as_wrapped(inline_segment):
+        if isinstance(item, TextSegment):
+            if text_segment is None:
+                text_segment = item
+            else:
+                yield ScoreSegment(
+                    text_segment=text_segment,
+                    left_parents=left_parents,
+                    right_parents=right_parents,
+                    text_tokens=[],
+                    score=0,
+                )
+                text_segment = item
+                left_parents = []
+                right_parents = []
+
+        elif isinstance(item, tuple):
+            child_inline_segment, orientation = item
+            if orientation == _Orientation.UP:
+                if text_segment is not None:
+                    yield ScoreSegment(
+                        text_segment=text_segment,
+                        left_parents=left_parents,
+                        right_parents=right_parents,
+                        text_tokens=[],
+                        score=0,
+                    )
+                    text_segment = None
+                    left_parents = []
+                    right_parents = []
+                left_parents.append(child_inline_segment)
+
+            elif orientation == _Orientation.DOWN:
+                if text_segment is None:
+                    left_parents.clear()
+                else:
+                    right_parents.append(child_inline_segment)
+
+    if text_segment is not None:
+        yield ScoreSegment(
+            text_segment=text_segment,
+            left_parents=left_parents,
+            right_parents=right_parents,
+            text_tokens=[],
+            score=0,
+        )
+
+
+class _Orientation(Enum):
+    DOWN = auto()
+    UP = auto()
+
+
+def _expand_as_wrapped(inline_segment: InlineSegment):
+    yield (inline_segment, _Orientation.UP)
+    for child in inline_segment.children:
+        if isinstance(child, InlineSegment):
+            yield from _expand_as_wrapped(child)
+        elif isinstance(child, TextSegment):
+            yield child
+    yield (inline_segment, _Orientation.DOWN)