Update app.py
Browse files
app.py
CHANGED
|
@@ -119,11 +119,11 @@ class TranslationQueue:
|
|
| 119 |
thread.start()
|
| 120 |
|
| 121 |
class TextChunker:
|
| 122 |
-
"""
|
| 123 |
|
| 124 |
@staticmethod
|
| 125 |
def split_text_smart(text: str, max_chunk_size: int = 400) -> List[str]:
|
| 126 |
-
"""
|
| 127 |
logger.info(f"[CHUNKER] Starting smart text splitting | Text length: {len(text)} chars | Max chunk size: {max_chunk_size}")
|
| 128 |
|
| 129 |
if len(text) <= max_chunk_size:
|
|
@@ -132,27 +132,27 @@ class TextChunker:
|
|
| 132 |
|
| 133 |
chunks = []
|
| 134 |
|
| 135 |
-
#
|
| 136 |
paragraphs = text.split('\n\n')
|
| 137 |
current_chunk = ""
|
| 138 |
|
| 139 |
for i, paragraph in enumerate(paragraphs):
|
| 140 |
logger.debug(f"[CHUNKER] Processing paragraph {i+1}/{len(paragraphs)} | Length: {len(paragraph)}")
|
| 141 |
|
| 142 |
-
#
|
| 143 |
if len(paragraph) > max_chunk_size:
|
| 144 |
-
#
|
| 145 |
if current_chunk.strip():
|
| 146 |
chunks.append(current_chunk.strip())
|
| 147 |
logger.debug(f"[CHUNKER] Added chunk from accumulated paragraphs | Length: {len(current_chunk.strip())}")
|
| 148 |
current_chunk = ""
|
| 149 |
|
| 150 |
-
#
|
| 151 |
sub_chunks = TextChunker._split_paragraph(paragraph, max_chunk_size)
|
| 152 |
chunks.extend(sub_chunks)
|
| 153 |
logger.debug(f"[CHUNKER] Split large paragraph into {len(sub_chunks)} sub-chunks")
|
| 154 |
else:
|
| 155 |
-
#
|
| 156 |
if len(current_chunk) + len(paragraph) + 2 > max_chunk_size:
|
| 157 |
if current_chunk.strip():
|
| 158 |
chunks.append(current_chunk.strip())
|
|
@@ -164,7 +164,7 @@ class TextChunker:
|
|
| 164 |
else:
|
| 165 |
current_chunk = paragraph
|
| 166 |
|
| 167 |
-
#
|
| 168 |
if current_chunk.strip():
|
| 169 |
chunks.append(current_chunk.strip())
|
| 170 |
logger.debug(f"[CHUNKER] Added final chunk | Length: {len(current_chunk.strip())}")
|
|
@@ -174,10 +174,10 @@ class TextChunker:
|
|
| 174 |
|
| 175 |
@staticmethod
|
| 176 |
def _split_paragraph(paragraph: str, max_chunk_size: int) -> List[str]:
|
| 177 |
-
"""
|
| 178 |
logger.debug(f"[CHUNKER] Splitting large paragraph | Length: {len(paragraph)}")
|
| 179 |
|
| 180 |
-
#
|
| 181 |
sentences = re.split(r'[.!?]+\s+', paragraph)
|
| 182 |
chunks = []
|
| 183 |
current_chunk = ""
|
|
@@ -186,12 +186,12 @@ class TextChunker:
|
|
| 186 |
if not sentence.strip():
|
| 187 |
continue
|
| 188 |
|
| 189 |
-
#
|
| 190 |
if not sentence.endswith(('.', '!', '?')):
|
| 191 |
sentence += '.'
|
| 192 |
|
| 193 |
if len(sentence) > max_chunk_size:
|
| 194 |
-
#
|
| 195 |
if current_chunk.strip():
|
| 196 |
chunks.append(current_chunk.strip())
|
| 197 |
current_chunk = ""
|
|
@@ -217,7 +217,7 @@ class TextChunker:
|
|
| 217 |
|
| 218 |
@staticmethod
|
| 219 |
def _split_by_comma(sentence: str, max_chunk_size: int) -> List[str]:
|
| 220 |
-
"""
|
| 221 |
logger.debug(f"[CHUNKER] Splitting long sentence by comma | Length: {len(sentence)}")
|
| 222 |
|
| 223 |
parts = sentence.split(', ')
|
|
@@ -226,12 +226,12 @@ class TextChunker:
|
|
| 226 |
|
| 227 |
for part in parts:
|
| 228 |
if len(part) > max_chunk_size:
|
| 229 |
-
#
|
| 230 |
if current_chunk.strip():
|
| 231 |
chunks.append(current_chunk.strip())
|
| 232 |
current_chunk = ""
|
| 233 |
|
| 234 |
-
#
|
| 235 |
while len(part) > max_chunk_size:
|
| 236 |
chunks.append(part[:max_chunk_size].strip())
|
| 237 |
part = part[max_chunk_size:].strip()
|
|
@@ -282,9 +282,9 @@ class MultilingualTranslator:
|
|
| 282 |
logger.error(f"[INIT] Error loading model: {e}")
|
| 283 |
raise
|
| 284 |
|
| 285 |
-
#
|
| 286 |
-
self.max_chunk_size = 350 #
|
| 287 |
-
self.min_chunk_overlap = 20 #
|
| 288 |
|
| 289 |
# Track translation progress
|
| 290 |
self.current_translation = {}
|
|
@@ -293,109 +293,79 @@ class MultilingualTranslator:
|
|
| 293 |
logger.info(f"[INIT] Translator initialized | Max chunk size: {self.max_chunk_size} chars")
|
| 294 |
|
| 295 |
def translate_chunk(self, text: str, source_lang: str, target_lang: str, chunk_index: int = 0, total_chunks: int = 1) -> str:
|
| 296 |
-
"""
|
| 297 |
try:
|
| 298 |
-
logger.info(f"[TRANSLATE] Starting chunk translation [{chunk_index+1}/{total_chunks}] | {source_lang}
|
| 299 |
-
|
| 300 |
-
# بررسی متن ورودی
|
| 301 |
-
if not text or not text.strip():
|
| 302 |
-
logger.warning(f"[TRANSLATE] Empty text provided for chunk {chunk_index+1}")
|
| 303 |
-
return ""
|
| 304 |
|
| 305 |
# Set source language for tokenizer
|
| 306 |
self.tokenizer.src_lang = source_lang
|
| 307 |
|
| 308 |
# Encode input
|
| 309 |
-
|
| 310 |
-
|
| 311 |
-
logger.debug(f"[TRANSLATE] Text encoded | Input tokens: {encoded.input_ids.shape[1]}")
|
| 312 |
-
except Exception as e:
|
| 313 |
-
logger.error(f"[TRANSLATE] Tokenization error for chunk {chunk_index+1}: {e}")
|
| 314 |
-
return f"[Tokenization Error: {str(e)}]"
|
| 315 |
|
| 316 |
# Generate translation with optimized parameters
|
| 317 |
start_time = time.time()
|
| 318 |
-
|
| 319 |
-
|
| 320 |
-
|
| 321 |
-
|
| 322 |
-
|
| 323 |
-
|
| 324 |
-
|
| 325 |
-
|
| 326 |
-
|
| 327 |
-
|
| 328 |
-
|
| 329 |
-
|
| 330 |
-
|
| 331 |
-
|
| 332 |
-
|
| 333 |
-
|
| 334 |
-
generation_time = time.time() - start_time
|
| 335 |
-
except Exception as e:
|
| 336 |
-
logger.error(f"[TRANSLATE] Generation error for chunk {chunk_index+1}: {e}")
|
| 337 |
-
return f"[Generation Error: {str(e)}]"
|
| 338 |
|
| 339 |
# Decode result
|
| 340 |
-
|
| 341 |
-
|
| 342 |
-
|
| 343 |
-
|
| 344 |
-
|
| 345 |
-
|
| 346 |
-
|
| 347 |
-
|
| 348 |
-
|
| 349 |
-
return f"[Empty Translation Result for chunk {chunk_index+1}]"
|
| 350 |
-
|
| 351 |
-
logger.info(f"[TRANSLATE] Chunk translation completed [{chunk_index+1}/{total_chunks}] | Generation time: {generation_time:.2f}s | Output length: {len(translation)} chars")
|
| 352 |
-
|
| 353 |
-
return translation
|
| 354 |
-
|
| 355 |
-
except Exception as e:
|
| 356 |
-
logger.error(f"[TRANSLATE] Decoding error for chunk {chunk_index+1}: {e}")
|
| 357 |
-
return f"[Decoding Error: {str(e)}]"
|
| 358 |
-
|
| 359 |
except Exception as e:
|
| 360 |
logger.error(f"[TRANSLATE] Chunk translation error [{chunk_index+1}/{total_chunks}]: {e}")
|
| 361 |
return f"[Translation Error: {str(e)}]"
|
| 362 |
-
|
| 363 |
-
|
| 364 |
|
| 365 |
def translate_text(self, text: str, source_lang: str, target_lang: str, session_id: str = None) -> Tuple[str, float, int]:
|
| 366 |
-
"""
|
| 367 |
start_time = time.time()
|
| 368 |
|
| 369 |
if not session_id:
|
| 370 |
session_id = hashlib.md5(f"{text[:100]}{time.time()}".encode()).hexdigest()[:8]
|
| 371 |
|
| 372 |
-
logger.info(f"[SESSION:{session_id}] Starting translation | {source_lang}
|
| 373 |
|
| 374 |
-
#
|
| 375 |
cached_result = self.cache.get(text, source_lang, target_lang)
|
| 376 |
-
if cached_result
|
| 377 |
logger.info(f"[SESSION:{session_id}] Translation completed from cache | Time: {time.time() - start_time:.2f}s")
|
| 378 |
return cached_result, time.time() - start_time, 1
|
| 379 |
|
| 380 |
try:
|
| 381 |
-
#
|
| 382 |
if len(text) <= self.max_chunk_size:
|
| 383 |
logger.info(f"[SESSION:{session_id}] Processing as short text")
|
| 384 |
translation = self.translate_chunk(text, source_lang, target_lang, 0, 1)
|
| 385 |
|
| 386 |
-
#
|
| 387 |
-
if not translation or not translation.strip():
|
| 388 |
-
logger.error(f"[SESSION:{session_id}] Empty translation received for short text")
|
| 389 |
-
return "Translation failed: Empty result received", time.time() - start_time, 0
|
| 390 |
-
|
| 391 |
-
# ذخیره در کش
|
| 392 |
self.cache.set(text, source_lang, target_lang, translation)
|
| 393 |
processing_time = time.time() - start_time
|
| 394 |
logger.info(f"[SESSION:{session_id}] Short text translation completed | Total time: {processing_time:.2f}s")
|
| 395 |
|
| 396 |
return translation, processing_time, 1
|
| 397 |
|
| 398 |
-
#
|
| 399 |
logger.info(f"[SESSION:{session_id}] Processing as long text - starting chunking")
|
| 400 |
chunks = TextChunker.split_text_smart(text, self.max_chunk_size)
|
| 401 |
logger.info(f"[SESSION:{session_id}] Text split into {len(chunks)} chunks")
|
|
@@ -410,16 +380,16 @@ class MultilingualTranslator:
|
|
| 410 |
'target_lang': target_lang
|
| 411 |
}
|
| 412 |
|
| 413 |
-
#
|
| 414 |
translated_chunks = []
|
| 415 |
for i, chunk in enumerate(chunks):
|
| 416 |
chunk_start_time = time.time()
|
| 417 |
logger.info(f"[SESSION:{session_id}] Starting chunk {i+1}/{len(chunks)} | Chunk length: {len(chunk)} chars")
|
| 418 |
|
| 419 |
-
#
|
| 420 |
chunk_translation = self.cache.get(chunk, source_lang, target_lang)
|
| 421 |
|
| 422 |
-
if not chunk_translation
|
| 423 |
# Estimate remaining time
|
| 424 |
if i > 0:
|
| 425 |
elapsed_time = time.time() - start_time
|
|
@@ -428,14 +398,8 @@ class MultilingualTranslator:
|
|
| 428 |
logger.info(f"[SESSION:{session_id}] Progress: {i}/{len(chunks)} | Avg time per chunk: {avg_time_per_chunk:.1f}s | Estimated remaining: {estimated_remaining:.1f}s")
|
| 429 |
|
| 430 |
chunk_translation = self.translate_chunk(chunk, source_lang, target_lang, i, len(chunks))
|
| 431 |
-
|
| 432 |
-
|
| 433 |
-
if not chunk_translation or not chunk_translation.strip():
|
| 434 |
-
logger.error(f"[SESSION:{session_id}] Empty translation for chunk {i+1}")
|
| 435 |
-
chunk_translation = f"[Translation failed for chunk {i+1}]"
|
| 436 |
-
else:
|
| 437 |
-
# ذخیره قسمت در کش
|
| 438 |
-
self.cache.set(chunk, source_lang, target_lang, chunk_translation)
|
| 439 |
|
| 440 |
chunk_time = time.time() - chunk_start_time
|
| 441 |
logger.info(f"[SESSION:{session_id}] Chunk {i+1}/{len(chunks)} translated in {chunk_time:.2f}s")
|
|
@@ -449,21 +413,15 @@ class MultilingualTranslator:
|
|
| 449 |
if session_id in self.current_translation:
|
| 450 |
self.current_translation[session_id]['completed_chunks'] = i + 1
|
| 451 |
|
| 452 |
-
#
|
| 453 |
if i < len(chunks) - 1:
|
| 454 |
time.sleep(0.1)
|
| 455 |
|
| 456 |
-
#
|
| 457 |
logger.info(f"[SESSION:{session_id}] Combining translated chunks")
|
| 458 |
final_translation = self._combine_translations(translated_chunks, text)
|
| 459 |
|
| 460 |
-
#
|
| 461 |
-
if not final_translation or not final_translation.strip():
|
| 462 |
-
logger.error(f"[SESSION:{session_id}] Final translation is empty after combining chunks")
|
| 463 |
-
processing_time = time.time() - start_time
|
| 464 |
-
return "Translation failed: Empty result after processing", processing_time, len(chunks)
|
| 465 |
-
|
| 466 |
-
# ذخیره نتیجه نهایی در کش
|
| 467 |
self.cache.set(text, source_lang, target_lang, final_translation)
|
| 468 |
|
| 469 |
processing_time = time.time() - start_time
|
|
@@ -481,8 +439,6 @@ class MultilingualTranslator:
|
|
| 481 |
with self.translation_lock:
|
| 482 |
self.current_translation.pop(session_id, None)
|
| 483 |
return f"Translation error: {str(e)}", time.time() - start_time, 0
|
| 484 |
-
|
| 485 |
-
|
| 486 |
|
| 487 |
def get_translation_progress(self, session_id: str) -> Dict:
|
| 488 |
"""Get current translation progress"""
|
|
@@ -509,9 +465,8 @@ class MultilingualTranslator:
|
|
| 509 |
}
|
| 510 |
|
| 511 |
def _combine_translations(self, translated_chunks: List[str], original_text: str) -> str:
|
| 512 |
-
"""
|
| 513 |
if not translated_chunks:
|
| 514 |
-
logger.warning("[COMBINER] No chunks to combine")
|
| 515 |
return ""
|
| 516 |
|
| 517 |
if len(translated_chunks) == 1:
|
|
@@ -519,31 +474,23 @@ class MultilingualTranslator:
|
|
| 519 |
|
| 520 |
logger.debug(f"[COMBINER] Combining {len(translated_chunks)} translated chunks")
|
| 521 |
|
| 522 |
-
#
|
| 523 |
-
valid_chunks = []
|
| 524 |
-
for i, chunk in enumerate(translated_chunks):
|
| 525 |
-
chunk = chunk.strip() if chunk else ""
|
| 526 |
-
|
| 527 |
-
# حذف قسمتهایی که پیام خطا هستند
|
| 528 |
-
if chunk and not chunk.startswith('[') and not chunk.startswith('Translation Error'):
|
| 529 |
-
valid_chunks.append(chunk)
|
| 530 |
-
else:
|
| 531 |
-
logger.warning(f"[COMBINER] Skipping invalid chunk {i}: {chunk[:50] if chunk else 'Empty'}")
|
| 532 |
-
|
| 533 |
-
if not valid_chunks:
|
| 534 |
-
logger.error("[COMBINER] No valid chunks found")
|
| 535 |
-
return "Translation failed: No valid content produced"
|
| 536 |
-
|
| 537 |
-
# ترکیب قسمتها با در نظر گیری ساختار اصلی متن
|
| 538 |
combined = []
|
| 539 |
|
| 540 |
-
for i, chunk in enumerate(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 541 |
if i > 0 and combined:
|
| 542 |
-
#
|
| 543 |
-
if not combined[-1].rstrip().endswith(('.', '!', '?', ':', '
|
| 544 |
combined[-1] += '.'
|
| 545 |
|
| 546 |
-
#
|
| 547 |
if '\n\n' in original_text:
|
| 548 |
combined.append('\n\n' + chunk)
|
| 549 |
else:
|
|
@@ -553,9 +500,9 @@ class MultilingualTranslator:
|
|
| 553 |
|
| 554 |
result = ''.join(combined)
|
| 555 |
|
| 556 |
-
#
|
| 557 |
-
result = re.sub(r'\s+', ' ', result) #
|
| 558 |
-
result = re.sub(r'\.+', '.', result) #
|
| 559 |
result = result.strip()
|
| 560 |
|
| 561 |
logger.debug(f"[COMBINER] Combined translation length: {len(result)} chars")
|
|
@@ -570,7 +517,6 @@ class MultilingualTranslator:
|
|
| 570 |
text, source_lang, target_lang, session_id
|
| 571 |
)
|
| 572 |
|
| 573 |
-
|
| 574 |
# Language mappings for M2M100 model
|
| 575 |
LANGUAGE_MAP = {
|
| 576 |
"English": "en",
|
|
@@ -707,7 +653,7 @@ async def api_translate(request: TranslationRequest):
|
|
| 707 |
# Alternative endpoint for form data (compatibility with WordPress)
|
| 708 |
@app.post("/api/translate/form")
|
| 709 |
async def api_translate_form(request: Request):
|
| 710 |
-
"""Non-blocking translation endpoint"""
|
| 711 |
try:
|
| 712 |
form_data = await request.form()
|
| 713 |
text = form_data.get("text", "")
|
|
@@ -727,13 +673,15 @@ async def api_translate_form(request: Request):
|
|
| 727 |
logger.info(f"[FORM API] Translation request | {source_lang} → {target_lang} | Length: {len(text)} chars")
|
| 728 |
|
| 729 |
if not text.strip():
|
| 730 |
-
|
|
|
|
| 731 |
|
| 732 |
source_code = LANGUAGE_MAP.get(source_lang)
|
| 733 |
target_code = LANGUAGE_MAP.get(target_lang)
|
| 734 |
|
| 735 |
if not source_code or not target_code:
|
| 736 |
-
|
|
|
|
| 737 |
|
| 738 |
# Generate session ID for tracking
|
| 739 |
session_id = hashlib.md5(f"{text[:100]}{time.time()}".encode()).hexdigest()[:8]
|
|
@@ -762,6 +710,16 @@ async def api_translate_form(request: Request):
|
|
| 762 |
text, source_code, target_code, session_id
|
| 763 |
)
|
| 764 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 765 |
return {
|
| 766 |
"translation": translation,
|
| 767 |
"source_language": source_lang,
|
|
@@ -774,7 +732,7 @@ async def api_translate_form(request: Request):
|
|
| 774 |
}
|
| 775 |
except Exception as e:
|
| 776 |
logger.error(f"[FORM API] Translation error: {str(e)}")
|
| 777 |
-
|
| 778 |
|
| 779 |
@app.get("/api/progress/{session_id}")
|
| 780 |
async def get_translation_progress(session_id: str):
|
|
@@ -900,7 +858,7 @@ async def get_server_status():
|
|
| 900 |
if active_sessions or background_tasks_count > 0:
|
| 901 |
if active_sessions:
|
| 902 |
latest_session = active_sessions[-1]
|
| 903 |
-
message = f"Processing chunk {latest_session['completed_chunks']}/{latest_session['total_chunks']} | {latest_session['source_lang']}
|
| 904 |
else:
|
| 905 |
message = f"{background_tasks_count} translation(s) in background queue"
|
| 906 |
|
|
@@ -927,7 +885,7 @@ async def get_server_status():
|
|
| 927 |
return {
|
| 928 |
"has_active_translation": True,
|
| 929 |
"status": "processing",
|
| 930 |
-
"message": f"Processing chunk {latest_session['completed_chunks']}/{latest_session['total_chunks']} | {latest_session['source_lang']}
|
| 931 |
"session_data": latest_session
|
| 932 |
}
|
| 933 |
else:
|
|
|
|
| 119 |
thread.start()
|
| 120 |
|
| 121 |
class TextChunker:
|
| 122 |
+
"""کلاس برای تقسیم متن طولانی به بخش‌های کوچکتر"""
|
| 123 |
|
| 124 |
@staticmethod
|
| 125 |
def split_text_smart(text: str, max_chunk_size: int = 400) -> List[str]:
|
| 126 |
+
"""تقسیم هوشمند متن بر اساس جملات Ùˆ پاراگراÙ‌ها"""
|
| 127 |
logger.info(f"[CHUNKER] Starting smart text splitting | Text length: {len(text)} chars | Max chunk size: {max_chunk_size}")
|
| 128 |
|
| 129 |
if len(text) <= max_chunk_size:
|
|
|
|
| 132 |
|
| 133 |
chunks = []
|
| 134 |
|
| 135 |
+
# تقسیم بر اساس پاراگراÙ‌ها
|
| 136 |
paragraphs = text.split('\n\n')
|
| 137 |
current_chunk = ""
|
| 138 |
|
| 139 |
for i, paragraph in enumerate(paragraphs):
|
| 140 |
logger.debug(f"[CHUNKER] Processing paragraph {i+1}/{len(paragraphs)} | Length: {len(paragraph)}")
|
| 141 |
|
| 142 |
+
# اگر پاراگرا٠خودش بزرگ است، آن را تقسیم کن
|
| 143 |
if len(paragraph) > max_chunk_size:
|
| 144 |
+
# ذخیره قسمت ÙØ¹Ù„ÛŒ اگر وجود دارد
|
| 145 |
if current_chunk.strip():
|
| 146 |
chunks.append(current_chunk.strip())
|
| 147 |
logger.debug(f"[CHUNKER] Added chunk from accumulated paragraphs | Length: {len(current_chunk.strip())}")
|
| 148 |
current_chunk = ""
|
| 149 |
|
| 150 |
+
# تقسیم پاراگرا٠بزرگ
|
| 151 |
sub_chunks = TextChunker._split_paragraph(paragraph, max_chunk_size)
|
| 152 |
chunks.extend(sub_chunks)
|
| 153 |
logger.debug(f"[CHUNKER] Split large paragraph into {len(sub_chunks)} sub-chunks")
|
| 154 |
else:
|
| 155 |
+
# بررسی اینکه آیا اضاÙÙ‡ کردن این پاراگرا٠از ØØ¯ تجاوز می‌کند
|
| 156 |
if len(current_chunk) + len(paragraph) + 2 > max_chunk_size:
|
| 157 |
if current_chunk.strip():
|
| 158 |
chunks.append(current_chunk.strip())
|
|
|
|
| 164 |
else:
|
| 165 |
current_chunk = paragraph
|
| 166 |
|
| 167 |
+
# اضاÙÙ‡ کردن آخرین قسمت
|
| 168 |
if current_chunk.strip():
|
| 169 |
chunks.append(current_chunk.strip())
|
| 170 |
logger.debug(f"[CHUNKER] Added final chunk | Length: {len(current_chunk.strip())}")
|
|
|
|
| 174 |
|
| 175 |
@staticmethod
|
| 176 |
def _split_paragraph(paragraph: str, max_chunk_size: int) -> List[str]:
|
| 177 |
+
"""تقسیم پاراگرا٠بزرگ به جملات"""
|
| 178 |
logger.debug(f"[CHUNKER] Splitting large paragraph | Length: {len(paragraph)}")
|
| 179 |
|
| 180 |
+
# تقسیم بر اساس جملات
|
| 181 |
sentences = re.split(r'[.!?]+\s+', paragraph)
|
| 182 |
chunks = []
|
| 183 |
current_chunk = ""
|
|
|
|
| 186 |
if not sentence.strip():
|
| 187 |
continue
|
| 188 |
|
| 189 |
+
# اضاÙÙ‡ کردن علامت نقطه اگر ØØ°Ù شده
|
| 190 |
if not sentence.endswith(('.', '!', '?')):
|
| 191 |
sentence += '.'
|
| 192 |
|
| 193 |
if len(sentence) > max_chunk_size:
|
| 194 |
+
# جمله خودش خیلی بلند است - تقسیم بر اساس کاما
|
| 195 |
if current_chunk.strip():
|
| 196 |
chunks.append(current_chunk.strip())
|
| 197 |
current_chunk = ""
|
|
|
|
| 217 |
|
| 218 |
@staticmethod
|
| 219 |
def _split_by_comma(sentence: str, max_chunk_size: int) -> List[str]:
|
| 220 |
+
"""تقسیم جمله طولانی بر اساس کاما"""
|
| 221 |
logger.debug(f"[CHUNKER] Splitting long sentence by comma | Length: {len(sentence)}")
|
| 222 |
|
| 223 |
parts = sentence.split(', ')
|
|
|
|
| 226 |
|
| 227 |
for part in parts:
|
| 228 |
if len(part) > max_chunk_size:
|
| 229 |
+
# قسمت خودش خیلی بلند است - تقسیم اجباری
|
| 230 |
if current_chunk.strip():
|
| 231 |
chunks.append(current_chunk.strip())
|
| 232 |
current_chunk = ""
|
| 233 |
|
| 234 |
+
# تقسیم اجباری بر اساس طول
|
| 235 |
while len(part) > max_chunk_size:
|
| 236 |
chunks.append(part[:max_chunk_size].strip())
|
| 237 |
part = part[max_chunk_size:].strip()
|
|
|
|
| 282 |
logger.error(f"[INIT] Error loading model: {e}")
|
| 283 |
raise
|
| 284 |
|
| 285 |
+
# تنظیمات بهینه برای ترجمه متن‌های بلند
|
| 286 |
+
self.max_chunk_size = 350 # ØØ¯Ø§Ú©Ø«Ø± طول هر قسمت
|
| 287 |
+
self.min_chunk_overlap = 20 # همپوشانی بین قسمت‌ها
|
| 288 |
|
| 289 |
# Track translation progress
|
| 290 |
self.current_translation = {}
|
|
|
|
| 293 |
logger.info(f"[INIT] Translator initialized | Max chunk size: {self.max_chunk_size} chars")
|
| 294 |
|
| 295 |
def translate_chunk(self, text: str, source_lang: str, target_lang: str, chunk_index: int = 0, total_chunks: int = 1) -> str:
|
| 296 |
+
"""ترجمه یک قسمت کوچک از متن"""
|
| 297 |
try:
|
| 298 |
+
logger.info(f"[TRANSLATE] Starting chunk translation [{chunk_index+1}/{total_chunks}] | {source_lang} → {target_lang} | Length: {len(text)} chars")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 299 |
|
| 300 |
# Set source language for tokenizer
|
| 301 |
self.tokenizer.src_lang = source_lang
|
| 302 |
|
| 303 |
# Encode input
|
| 304 |
+
encoded = self.tokenizer(text, return_tensors="pt", truncation=True, max_length=512).to(self.device)
|
| 305 |
+
logger.debug(f"[TRANSLATE] Text encoded | Input tokens: {encoded.input_ids.shape[1]}")
|
|
|
|
|
|
|
|
|
|
|
|
|
| 306 |
|
| 307 |
# Generate translation with optimized parameters
|
| 308 |
start_time = time.time()
|
| 309 |
+
generated_tokens = self.model.generate(
|
| 310 |
+
**encoded,
|
| 311 |
+
forced_bos_token_id=self.tokenizer.get_lang_id(target_lang),
|
| 312 |
+
max_length=1024, # Ø§ÙØ²Ø§ÛŒØ´ طول خروجی
|
| 313 |
+
min_length=10, # ØØ¯Ø§Ù‚Ù„ طول خروجی
|
| 314 |
+
num_beams=5, # Ø§ÙØ²Ø§ÛŒØ´ تعداد beam ها برای Ú©ÛŒÙیت بهتر
|
| 315 |
+
early_stopping=True,
|
| 316 |
+
no_repeat_ngram_size=3, # جلوگیری از تکرار
|
| 317 |
+
length_penalty=1.0, # تنظیم جریمه طول
|
| 318 |
+
repetition_penalty=1.2, # جلوگیری از تکرار کلمات
|
| 319 |
+
do_sample=False, # Ø§Ø³ØªÙØ§Ø¯Ù‡ از روش قطعی
|
| 320 |
+
temperature=0.7, # کنترل تنوع
|
| 321 |
+
pad_token_id=self.tokenizer.pad_token_id,
|
| 322 |
+
eos_token_id=self.tokenizer.eos_token_id
|
| 323 |
+
)
|
| 324 |
+
generation_time = time.time() - start_time
|
|
|
|
|
|
|
|
|
|
|
|
|
| 325 |
|
| 326 |
# Decode result
|
| 327 |
+
translation = self.tokenizer.batch_decode(generated_tokens, skip_special_tokens=True)[0]
|
| 328 |
+
|
| 329 |
+
# پاک‌سازی ترجمه از کاراکترهای اضاÙÛŒ
|
| 330 |
+
translation = translation.strip()
|
| 331 |
+
|
| 332 |
+
logger.info(f"[TRANSLATE] Chunk translation completed [{chunk_index+1}/{total_chunks}] | Generation time: {generation_time:.2f}s | Output length: {len(translation)} chars")
|
| 333 |
+
|
| 334 |
+
return translation
|
| 335 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 336 |
except Exception as e:
|
| 337 |
logger.error(f"[TRANSLATE] Chunk translation error [{chunk_index+1}/{total_chunks}]: {e}")
|
| 338 |
return f"[Translation Error: {str(e)}]"
|
|
|
|
|
|
|
| 339 |
|
| 340 |
def translate_text(self, text: str, source_lang: str, target_lang: str, session_id: str = None) -> Tuple[str, float, int]:
|
| 341 |
+
"""ترجمه متن با پشتیبانی از متن‌های طولانی Ùˆ لاگ‌های Ù…ÙØµÙ„"""
|
| 342 |
start_time = time.time()
|
| 343 |
|
| 344 |
if not session_id:
|
| 345 |
session_id = hashlib.md5(f"{text[:100]}{time.time()}".encode()).hexdigest()[:8]
|
| 346 |
|
| 347 |
+
logger.info(f"[SESSION:{session_id}] Starting translation | {source_lang} → {target_lang} | Text length: {len(text)} chars")
|
| 348 |
|
| 349 |
+
# بررسی کش برای کل متن
|
| 350 |
cached_result = self.cache.get(text, source_lang, target_lang)
|
| 351 |
+
if cached_result:
|
| 352 |
logger.info(f"[SESSION:{session_id}] Translation completed from cache | Time: {time.time() - start_time:.2f}s")
|
| 353 |
return cached_result, time.time() - start_time, 1
|
| 354 |
|
| 355 |
try:
|
| 356 |
+
# اگر متن کوتاه است، مستقیماً ترجمه کن
|
| 357 |
if len(text) <= self.max_chunk_size:
|
| 358 |
logger.info(f"[SESSION:{session_id}] Processing as short text")
|
| 359 |
translation = self.translate_chunk(text, source_lang, target_lang, 0, 1)
|
| 360 |
|
| 361 |
+
# ذخیره در کش
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 362 |
self.cache.set(text, source_lang, target_lang, translation)
|
| 363 |
processing_time = time.time() - start_time
|
| 364 |
logger.info(f"[SESSION:{session_id}] Short text translation completed | Total time: {processing_time:.2f}s")
|
| 365 |
|
| 366 |
return translation, processing_time, 1
|
| 367 |
|
| 368 |
+
# تقسیم متن طولانی به قسمت‌های کوچکتر
|
| 369 |
logger.info(f"[SESSION:{session_id}] Processing as long text - starting chunking")
|
| 370 |
chunks = TextChunker.split_text_smart(text, self.max_chunk_size)
|
| 371 |
logger.info(f"[SESSION:{session_id}] Text split into {len(chunks)} chunks")
|
|
|
|
| 380 |
'target_lang': target_lang
|
| 381 |
}
|
| 382 |
|
| 383 |
+
# ترجمه هر قسمت
|
| 384 |
translated_chunks = []
|
| 385 |
for i, chunk in enumerate(chunks):
|
| 386 |
chunk_start_time = time.time()
|
| 387 |
logger.info(f"[SESSION:{session_id}] Starting chunk {i+1}/{len(chunks)} | Chunk length: {len(chunk)} chars")
|
| 388 |
|
| 389 |
+
# بررسی کش برای هر قسمت
|
| 390 |
chunk_translation = self.cache.get(chunk, source_lang, target_lang)
|
| 391 |
|
| 392 |
+
if not chunk_translation:
|
| 393 |
# Estimate remaining time
|
| 394 |
if i > 0:
|
| 395 |
elapsed_time = time.time() - start_time
|
|
|
|
| 398 |
logger.info(f"[SESSION:{session_id}] Progress: {i}/{len(chunks)} | Avg time per chunk: {avg_time_per_chunk:.1f}s | Estimated remaining: {estimated_remaining:.1f}s")
|
| 399 |
|
| 400 |
chunk_translation = self.translate_chunk(chunk, source_lang, target_lang, i, len(chunks))
|
| 401 |
+
# ذخیره قسمت در کش
|
| 402 |
+
self.cache.set(chunk, source_lang, target_lang, chunk_translation)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 403 |
|
| 404 |
chunk_time = time.time() - chunk_start_time
|
| 405 |
logger.info(f"[SESSION:{session_id}] Chunk {i+1}/{len(chunks)} translated in {chunk_time:.2f}s")
|
|
|
|
| 413 |
if session_id in self.current_translation:
|
| 414 |
self.current_translation[session_id]['completed_chunks'] = i + 1
|
| 415 |
|
| 416 |
+
# Ú©Ù…ÛŒ Ø§Ø³ØªØ±Ø§ØØª بین ترجمه‌ها برای جلوگیری از بارگذاری زیاد
|
| 417 |
if i < len(chunks) - 1:
|
| 418 |
time.sleep(0.1)
|
| 419 |
|
| 420 |
+
# ترکیب قسمت‌های ترجمه شده
|
| 421 |
logger.info(f"[SESSION:{session_id}] Combining translated chunks")
|
| 422 |
final_translation = self._combine_translations(translated_chunks, text)
|
| 423 |
|
| 424 |
+
# ��°Ø®ÛŒØ±Ù‡ نتیجه نهایی در Ú©Ø´
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 425 |
self.cache.set(text, source_lang, target_lang, final_translation)
|
| 426 |
|
| 427 |
processing_time = time.time() - start_time
|
|
|
|
| 439 |
with self.translation_lock:
|
| 440 |
self.current_translation.pop(session_id, None)
|
| 441 |
return f"Translation error: {str(e)}", time.time() - start_time, 0
|
|
|
|
|
|
|
| 442 |
|
| 443 |
def get_translation_progress(self, session_id: str) -> Dict:
|
| 444 |
"""Get current translation progress"""
|
|
|
|
| 465 |
}
|
| 466 |
|
| 467 |
def _combine_translations(self, translated_chunks: List[str], original_text: str) -> str:
|
| 468 |
+
"""ترکیب قسمت‌های ترجمه شده به یک متن یکپارچه"""
|
| 469 |
if not translated_chunks:
|
|
|
|
| 470 |
return ""
|
| 471 |
|
| 472 |
if len(translated_chunks) == 1:
|
|
|
|
| 474 |
|
| 475 |
logger.debug(f"[COMBINER] Combining {len(translated_chunks)} translated chunks")
|
| 476 |
|
| 477 |
+
# ترکیب قسمت‌ها با در نظر گیری ساختار اصلی متن
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 478 |
combined = []
|
| 479 |
|
| 480 |
+
for i, chunk in enumerate(translated_chunks):
|
| 481 |
+
# پاک‌سازی قسمت
|
| 482 |
+
chunk = chunk.strip()
|
| 483 |
+
|
| 484 |
+
if not chunk:
|
| 485 |
+
continue
|
| 486 |
+
|
| 487 |
+
# اضاÙÙ‡ کردن ÙØ§ØµÙ„Ù‡ مناسب بین قسمت‌ها
|
| 488 |
if i > 0 and combined:
|
| 489 |
+
# اگر قسمت قبلی با نقطه تمام نمی‌شود، نقطه اضاÙÙ‡ Ú©Ù†
|
| 490 |
+
if not combined[-1].rstrip().endswith(('.', '!', '?', ':', 'Ø›', '.')):
|
| 491 |
combined[-1] += '.'
|
| 492 |
|
| 493 |
+
# بررسی اینکه آیا نیاز به پاراگرا٠جدید داریم
|
| 494 |
if '\n\n' in original_text:
|
| 495 |
combined.append('\n\n' + chunk)
|
| 496 |
else:
|
|
|
|
| 500 |
|
| 501 |
result = ''.join(combined)
|
| 502 |
|
| 503 |
+
# پاک‌سازی نهایی
|
| 504 |
+
result = re.sub(r'\s+', ' ', result) # ØØ°Ù ÙØ§ØµÙ„ه‌های اضاÙÛŒ
|
| 505 |
+
result = re.sub(r'\.+', '.', result) # ØØ°Ù نقطه‌های تکراری
|
| 506 |
result = result.strip()
|
| 507 |
|
| 508 |
logger.debug(f"[COMBINER] Combined translation length: {len(result)} chars")
|
|
|
|
| 517 |
text, source_lang, target_lang, session_id
|
| 518 |
)
|
| 519 |
|
|
|
|
| 520 |
# Language mappings for M2M100 model
|
| 521 |
LANGUAGE_MAP = {
|
| 522 |
"English": "en",
|
|
|
|
| 653 |
# Alternative endpoint for form data (compatibility with WordPress)
|
| 654 |
@app.post("/api/translate/form")
|
| 655 |
async def api_translate_form(request: Request):
|
| 656 |
+
"""Non-blocking translation endpoint with enhanced error handling"""
|
| 657 |
try:
|
| 658 |
form_data = await request.form()
|
| 659 |
text = form_data.get("text", "")
|
|
|
|
| 673 |
logger.info(f"[FORM API] Translation request | {source_lang} → {target_lang} | Length: {len(text)} chars")
|
| 674 |
|
| 675 |
if not text.strip():
|
| 676 |
+
logger.error("[FORM API] No text provided")
|
| 677 |
+
return {"status": "error", "message": "No text provided"}
|
| 678 |
|
| 679 |
source_code = LANGUAGE_MAP.get(source_lang)
|
| 680 |
target_code = LANGUAGE_MAP.get(target_lang)
|
| 681 |
|
| 682 |
if not source_code or not target_code:
|
| 683 |
+
logger.error(f"[FORM API] Invalid language codes: {source_lang} -> {target_lang}")
|
| 684 |
+
return {"status": "error", "message": "Invalid language codes"}
|
| 685 |
|
| 686 |
# Generate session ID for tracking
|
| 687 |
session_id = hashlib.md5(f"{text[:100]}{time.time()}".encode()).hexdigest()[:8]
|
|
|
|
| 710 |
text, source_code, target_code, session_id
|
| 711 |
)
|
| 712 |
|
| 713 |
+
# بررسی محتوای ترجمه
|
| 714 |
+
if not translation or not translation.strip() or translation.startswith("Translation error"):
|
| 715 |
+
logger.error(f"[FORM API] Invalid translation result: {translation[:100] if translation else 'None'}")
|
| 716 |
+
return {
|
| 717 |
+
"status": "error",
|
| 718 |
+
"message": "Translation failed - empty or invalid result",
|
| 719 |
+
"session_id": session_id
|
| 720 |
+
}
|
| 721 |
+
|
| 722 |
+
logger.info(f"[FORM API] Translation successful | Length: {len(translation)} chars")
|
| 723 |
return {
|
| 724 |
"translation": translation,
|
| 725 |
"source_language": source_lang,
|
|
|
|
| 732 |
}
|
| 733 |
except Exception as e:
|
| 734 |
logger.error(f"[FORM API] Translation error: {str(e)}")
|
| 735 |
+
return {"status": "error", "message": f"Translation error: {str(e)}"}
|
| 736 |
|
| 737 |
@app.get("/api/progress/{session_id}")
|
| 738 |
async def get_translation_progress(session_id: str):
|
|
|
|
| 858 |
if active_sessions or background_tasks_count > 0:
|
| 859 |
if active_sessions:
|
| 860 |
latest_session = active_sessions[-1]
|
| 861 |
+
message = f"Processing chunk {latest_session['completed_chunks']}/{latest_session['total_chunks']} | {latest_session['source_lang']} → {latest_session['target_lang']}"
|
| 862 |
else:
|
| 863 |
message = f"{background_tasks_count} translation(s) in background queue"
|
| 864 |
|
|
|
|
| 885 |
return {
|
| 886 |
"has_active_translation": True,
|
| 887 |
"status": "processing",
|
| 888 |
+
"message": f"Processing chunk {latest_session['completed_chunks']}/{latest_session['total_chunks']} | {latest_session['source_lang']} → {latest_session['target_lang']}",
|
| 889 |
"session_data": latest_session
|
| 890 |
}
|
| 891 |
else:
|