ai-saas / saas-search

Commit 14e67b717157040d906b1e33098f59e1f9d66ba3

Authored by tangwang
1 parent 294c3d0a

分句后的 batching 现在是“先全量分句，再按 segment 总数按模型 batch_size 

推理”，不再是先按原始输入条数切块。也就是说，如果 100 条请求分句后变成
150 个 segments，batch_size=64 时会按 64 + 64 + 22
三批推理，推理完再按原始分句计划合并并还原成 100 条返回。这个改动在
local_seq2seq.py (line 241) 和 local_ctranslate2.py (line 391)。

日志这边也补上了两层你要的关键信息：

分句摘要日志：Translation segmentation
summary，会打印输入条数、非空条数、发生分句的输入数、总 segments
数、当前 batch_size、每条输入分成多少段的统计，见 local_seq2seq.py (line
216) 和 local_ctranslate2.py (line 366)。
每个预测批次日志：Translation inference
batch，会打印第几批、总批数、该批 segment
数、长度统计、首条预览。CTranslate2 另外还会打印 Translation model batch
detail，补充 token 长度和 max_decoding_length，见 local_ctranslate2.py
(line 294)。
我也补了测试，覆盖了“分句后再
batching”和“日志中有分句摘要与每批推理日志”，在
test_translation_local_backends.py (line 358)。
Showing 10 changed files with 610 additions and 113 deletions   Show diff stats
Inline Side-by-side
CLAUDE.md
  Show/Hide comments   View file @14e67b7
... ... @@ -23,12 +23,24 @@ This is a **production-ready Multi-Tenant E-Commerce Search SaaS** platform spec
23 23  
24 24 ## Development Environment
25 25  
26   -**Required Environment Setup:** Use project root `activate.sh` (activates conda env `searchengine` and loads `.env`). On a new machine, set `CONDA_ROOT` if conda is not at default path.
  26 +**Required Environment Setup:** Default to the project venv via root `activate.sh` (activates `./.venv` and loads `.env`). Do not default to system `python3`/`pip3` for repo work.
27 27 ```bash
28   -# Optional on new machine: if conda is ~/anaconda3/bin/conda → export CONDA_ROOT=$HOME/anaconda3
29 28 source activate.sh
30 29 ```
31   -See `docs/QUICKSTART.md` §1.4–1.8 for first-time env creation and production credentials (venv: `./scripts/create_venv.sh`; conda: `conda env create -f environment.yml` or `pip install -r requirements.txt`).
  30 +See `docs/QUICKSTART.md` §1.4–1.8 for first-time env creation and production credentials (`./scripts/create_venv.sh` for the main venv).
  31 +
  32 +**Environment Resolution Rules:**
  33 +- Main app, backend, indexer, frontend, generic scripts, and most tests: `./.venv`
  34 +- Translator service runtime and local translation model tooling: `./.venv-translator`
  35 +- Embedding service runtime: `./.venv-embedding`
  36 +- Reranker service runtime: `./.venv-reranker`
  37 +- CN-CLIP service runtime: `./.venv-cnclip`
  38 +- Never assume the system interpreter reflects project dependencies; prefer `source activate.sh` or invoke the exact venv binary directly.
  39 +
  40 +**Operational Rule For Commands:**
  41 +- For repo-wide `pytest`, ad hoc Python scripts, and lightweight development commands, use the matching venv interpreter first.
  42 +- For isolated services, prefer the service scripts (`./scripts/start_translator.sh`, `./scripts/start_embedding_service.sh`, `./scripts/start_reranker.sh`, `./scripts/start_cnclip_service.sh`) because they already select the correct environment.
  43 +- If a dependency appears “missing”, check whether the command was run under the wrong venv before proposing installs.
32 44  
33 45 **Database Configuration:**
34 46 ```yaml
... ... @@ -49,12 +61,14 @@ password: P89cZHS5d7dFyc9R
49 61  
50 62 ### Environment Setup
51 63 ```bash
52   -# Activate environment (canonical: use activate.sh)
  64 +# Activate main project environment (canonical)
53 65 source activate.sh
54 66  
55 67 # First-time / new machine: create env and install deps
56   -./setup.sh # or: conda env create -f environment.yml
57   -# If pip-only: pip install -r requirements.txt
  68 +./setup.sh
  69 +# or:
  70 +./scripts/create_venv.sh
  71 +source activate.sh
58 72 ```
59 73  
60 74 ### Data Management
... ... @@ -83,12 +97,12 @@ python main.py serve --host 0.0.0.0 --port 6002 --reload
83 97 ### Testing
84 98 ```bash
85 99 # Run all tests
86   -python -m pytest tests/
  100 +pytest tests/
87 101  
88 102 # Run specific test types
89   -python -m pytest tests/unit/ # Unit tests
90   -python -m pytest tests/integration/ # Integration tests
91   -python -m pytest -m "api" # API tests only
  103 +pytest tests/unit/ # Unit tests
  104 +pytest tests/integration/ # Integration tests
  105 +pytest -m "api" # API tests only
92 106  
93 107 # Test search from command line
94 108 python main.py search "query" --tenant-id 1 --size 10
... ... @@ -602,4 +616,3 @@ python main.py search "query" --tenant-id 1 # Quick search test
602 616 8. **Multi-tenant Architecture**: Single index with `tenant_id` isolation
603 617 9. **Hybrid Search**: BM25 + vector similarity with configurable weighting
604 618 10. **Production Ready**: Health checks, monitoring, graceful degradation
605   -
... ...
api/translator_app.py
  Show/Hide comments   View file @14e67b7
... ... @@ -5,18 +5,24 @@ import logging
5 5 import os
6 6 import pathlib
7 7 import time
  8 +import uuid
8 9 from contextlib import asynccontextmanager
9 10 from functools import lru_cache
10 11 from logging.handlers import TimedRotatingFileHandler
11 12 from typing import List, Optional, Union
12 13  
13 14 import uvicorn
14   -from fastapi import FastAPI, HTTPException
  15 +from fastapi import FastAPI, HTTPException, Request
15 16 from fastapi.middleware.cors import CORSMiddleware
16 17 from fastapi.responses import JSONResponse
17 18 from pydantic import BaseModel, ConfigDict, Field
18 19  
19 20 from config.services_config import get_translation_config
  21 +from translation.logging_utils import (
  22 + TranslationRequestFilter,
  23 + bind_translation_request_id,
  24 + reset_translation_request_id,
  25 +)
20 26 from translation.service import TranslationService
21 27 from translation.settings import (
22 28 get_enabled_translation_models,
... ... @@ -33,7 +39,8 @@ def configure_translator_logging() -> None:
33 39  
34 40 log_level = os.getenv("LOG_LEVEL", "INFO").upper()
35 41 numeric_level = getattr(logging, log_level, logging.INFO)
36   - formatter = logging.Formatter("%(asctime)s - %(name)s - %(levelname)s - %(message)s")
  42 + formatter = logging.Formatter("%(asctime)s | reqid:%(reqid)s | %(name)s | %(levelname)s | %(message)s")
  43 + request_filter = TranslationRequestFilter()
37 44  
38 45 root_logger = logging.getLogger()
39 46 root_logger.setLevel(numeric_level)
... ... @@ -42,6 +49,7 @@ def configure_translator_logging() -> None:
42 49 console_handler = logging.StreamHandler()
43 50 console_handler.setLevel(numeric_level)
44 51 console_handler.setFormatter(formatter)
  52 + console_handler.addFilter(request_filter)
45 53 root_logger.addHandler(console_handler)
46 54  
47 55 file_handler = TimedRotatingFileHandler(
... ... @@ -53,6 +61,7 @@ def configure_translator_logging() -> None:
53 61 )
54 62 file_handler.setLevel(numeric_level)
55 63 file_handler.setFormatter(formatter)
  64 + file_handler.addFilter(request_filter)
56 65 root_logger.addHandler(file_handler)
57 66  
58 67 verbose_logger = logging.getLogger("translator.verbose")
... ... @@ -69,6 +78,7 @@ def configure_translator_logging() -> None:
69 78 )
70 79 verbose_handler.setLevel(numeric_level)
71 80 verbose_handler.setFormatter(formatter)
  81 + verbose_handler.addFilter(request_filter)
72 82 verbose_logger.addHandler(verbose_handler)
73 83  
74 84  
... ... @@ -178,6 +188,13 @@ def _result_preview(translated: Union[str, List[Optional[str]], None]) -> str:
178 188 return _text_preview(str(translated))
179 189  
180 190  
  191 +def _resolve_request_id(http_request: Request) -> str:
  192 + header_value = http_request.headers.get("X-Request-ID")
  193 + if header_value and header_value.strip():
  194 + return header_value.strip()[:32]
  195 + return str(uuid.uuid4())[:8]
  196 +
  197 +
181 198 def _translate_batch(
182 199 service: TranslationService,
183 200 raw_text: List[str],
... ... @@ -189,15 +206,11 @@ def _translate_batch(
189 206 ) -> List[Optional[str]]:
190 207 backend = service.get_backend(model)
191 208 logger.info(
192   - "Translation batch dispatch | model=%s scene=%s target_lang=%s source_lang=%s count=%s lengths=%s first_preview=%s supports_batch=%s",
193   - model,
194   - scene,
195   - target_lang,
196   - source_lang or "auto",
  209 + "Translation batch dispatch | execution=%s count=%s lengths=%s first_preview=%s",
  210 + "backend-batch" if getattr(backend, "supports_batch", False) else "per-item",
197 211 len(raw_text),
198 212 [len(str(item or "")) for item in raw_text],
199 213 _text_preview(raw_text[0] if raw_text else ""),
200   - bool(getattr(backend, "supports_batch", False)),
201 214 )
202 215 if getattr(backend, "supports_batch", False):
203 216 try:
... ... @@ -330,12 +343,13 @@ async def health_check():
330 343  
331 344  
332 345 @app.post("/translate", response_model=TranslationResponse)
333   -async def translate(request: TranslationRequest):
  346 +async def translate(request: TranslationRequest, http_request: Request):
334 347 _ensure_valid_text(request.text)
335 348  
336 349 if not request.target_lang:
337 350 raise HTTPException(status_code=400, detail="target_lang is required")
338 351  
  352 + _, request_token = bind_translation_request_id(_resolve_request_id(http_request))
339 353 request_started = time.perf_counter()
340 354 try:
341 355 service = get_translation_service()
... ... @@ -447,12 +461,14 @@ async def translate(request: TranslationRequest):
447 461 raise
448 462 except ValueError as e:
449 463 latency_ms = (time.perf_counter() - request_started) * 1000
450   - logger.warning("Translation validation error | error=%s latency_ms=%.2f", e, latency_ms, exc_info=True)
  464 + logger.warning("Translation validation error | error=%s latency_ms=%.2f", e, latency_ms)
451 465 raise HTTPException(status_code=400, detail=str(e)) from e
452 466 except Exception as e:
453 467 latency_ms = (time.perf_counter() - request_started) * 1000
454 468 logger.error("Translation error | error=%s latency_ms=%.2f", e, latency_ms, exc_info=True)
455 469 raise HTTPException(status_code=500, detail=f"Translation error: {str(e)}")
  470 + finally:
  471 + reset_translation_request_id(request_token)
456 472  
457 473  
458 474 @app.get("/")
... ...
tests/test_translation_local_backends.py
  Show/Hide comments   View file @14e67b7
  1 +import logging
  2 +
  3 +import pytest
1 4 import torch
2 5  
3 6 from translation.backends.local_seq2seq import MarianMTTranslationBackend, NLLBTranslationBackend
  7 +from translation.backends.local_ctranslate2 import NLLBCTranslate2TranslationBackend
4 8 from translation.service import TranslationService
5 9 from translation.text_splitter import compute_safe_input_token_limit, split_text_for_translation
6 10  
... ... @@ -44,11 +48,59 @@ class _FakeModel:
44 48 return [[42]]
45 49  
46 50  
  51 +class _FakeCT2Tokenizer:
  52 + def __init__(self, src_lang=None):
  53 + self.src_lang = src_lang
  54 + self.pad_token = "</s>"
  55 + self.eos_token = "</s>"
  56 + self.last_call = None
  57 +
  58 + def __call__(self, texts, **kwargs):
  59 + self.last_call = {"texts": list(texts), **kwargs}
  60 + return {"input_ids": [[1, 2, 3] for _ in texts]}
  61 +
  62 + def convert_ids_to_tokens(self, ids):
  63 + del ids
  64 + return ["tok_a", "tok_b", "tok_c"]
  65 +
  66 + def convert_tokens_to_ids(self, tokens):
  67 + if isinstance(tokens, list):
  68 + return [1 for _ in tokens]
  69 + return 1
  70 +
  71 + def decode(self, token_ids, skip_special_tokens=True):
  72 + del token_ids, skip_special_tokens
  73 + return "translated"
  74 +
  75 +
  76 +class _FakeCT2Result:
  77 + def __init__(self, tokens):
  78 + self.hypotheses = [tokens]
  79 +
  80 +
  81 +class _FakeCT2Translator:
  82 + def __init__(self):
  83 + self.last_translate_batch_kwargs = None
  84 +
  85 + def translate_batch(self, source_tokens, **kwargs):
  86 + self.last_translate_batch_kwargs = {"source_tokens": source_tokens, **kwargs}
  87 + target_prefix = kwargs.get("target_prefix") or []
  88 + return [
  89 + _FakeCT2Result((target_prefix[idx] or []) + ["translated_token"])
  90 + for idx, _ in enumerate(source_tokens)
  91 + ]
  92 +
  93 +
47 94 def _stub_load_model(self):
48 95 self.tokenizer = _FakeTokenizer()
49 96 self.seq2seq_model = _FakeModel()
50 97  
51 98  
  99 +def _stub_load_ct2_runtime(self):
  100 + self.tokenizer = _FakeCT2Tokenizer()
  101 + self.translator = _FakeCT2Translator()
  102 +
  103 +
52 104 def test_marian_language_validation(monkeypatch):
53 105 monkeypatch.setattr(MarianMTTranslationBackend, "_load_model", _stub_load_model)
54 106 backend = MarianMTTranslationBackend(
... ... @@ -68,12 +120,8 @@ def test_marian_language_validation(monkeypatch):
68 120 result = backend.translate("测试", source_lang="zh", target_lang="en")
69 121 assert result == "translated"
70 122  
71   - try:
  123 + with pytest.raises(ValueError, match="source languages"):
72 124 backend.translate("test", source_lang="en", target_lang="zh")
73   - except ValueError as exc:
74   - assert "source languages" in str(exc)
75   - else:
76   - raise AssertionError("Expected unsupported source language to raise")
77 125  
78 126  
79 127 def test_nllb_uses_src_lang_and_forced_bos(monkeypatch):
... ... @@ -97,6 +145,61 @@ def test_nllb_uses_src_lang_and_forced_bos(monkeypatch):
97 145 assert backend.seq2seq_model.last_generate_kwargs["forced_bos_token_id"] == 202
98 146  
99 147  
  148 +def test_nllb_accepts_finnish_short_code(monkeypatch):
  149 + monkeypatch.setattr(NLLBTranslationBackend, "_load_model", _stub_load_model)
  150 + backend = NLLBTranslationBackend(
  151 + name="nllb-200-distilled-600m",
  152 + model_id="facebook/nllb-200-distilled-600M",
  153 + model_dir="./models/translation/facebook/nllb-200-distilled-600M",
  154 + device="cpu",
  155 + torch_dtype="float32",
  156 + batch_size=1,
  157 + max_input_length=16,
  158 + max_new_tokens=16,
  159 + num_beams=1,
  160 + )
  161 +
  162 + result = backend.translate("test", source_lang="fi", target_lang="zh")
  163 +
  164 + assert result == "translated"
  165 + assert backend.tokenizer.src_lang == "fin_Latn"
  166 + assert backend.seq2seq_model.last_generate_kwargs["forced_bos_token_id"] == 202
  167 +
  168 +
  169 +def test_nllb_ctranslate2_accepts_finnish_short_code(monkeypatch):
  170 + created_tokenizers = []
  171 +
  172 + def _fake_from_pretrained(source, src_lang=None, **kwargs):
  173 + del source, kwargs
  174 + tokenizer = _FakeCT2Tokenizer(src_lang=src_lang)
  175 + created_tokenizers.append(tokenizer)
  176 + return tokenizer
  177 +
  178 + monkeypatch.setattr(NLLBCTranslate2TranslationBackend, "_load_runtime", _stub_load_ct2_runtime)
  179 + monkeypatch.setattr(
  180 + "translation.backends.local_ctranslate2.AutoTokenizer.from_pretrained",
  181 + _fake_from_pretrained,
  182 + )
  183 + backend = NLLBCTranslate2TranslationBackend(
  184 + name="nllb-200-distilled-600m",
  185 + model_id="facebook/nllb-200-distilled-600M",
  186 + model_dir="./models/translation/facebook/nllb-200-distilled-600M",
  187 + device="cpu",
  188 + torch_dtype="float32",
  189 + batch_size=1,
  190 + max_input_length=16,
  191 + max_new_tokens=16,
  192 + num_beams=1,
  193 + )
  194 +
  195 + result = backend.translate("test", source_lang="fi", target_lang="zh")
  196 +
  197 + assert result == "translated"
  198 + assert len(created_tokenizers) == 1
  199 + assert created_tokenizers[0].src_lang == "fin_Latn"
  200 + assert backend.translator.last_translate_batch_kwargs["target_prefix"] == [["zho_Hans"]]
  201 +
  202 +
100 203 def test_translation_service_preloads_enabled_backends(monkeypatch):
101 204 created = []
102 205  
... ... @@ -245,7 +348,71 @@ def test_local_backend_splits_oversized_text_before_translation():
245 348 result = backend.translate(text, source_lang="en", target_lang="zh")
246 349  
247 350 assert result is not None
248   - assert len(backend.translated_batches) == 1
249   - assert len(backend.translated_batches[0]) >= 2
250   - assert all(len(piece) <= 16 for piece in backend.translated_batches[0])
251   - assert result == "".join(f"<{piece.strip()}>" for piece in backend.translated_batches[0])
  351 + all_segments = [piece for batch in backend.translated_batches for piece in batch]
  352 + assert len(all_segments) >= 2
  353 + assert all(len(batch) <= backend.batch_size for batch in backend.translated_batches)
  354 + assert all(len(piece) <= 16 for piece in all_segments)
  355 + assert result == "".join(f"<{piece.strip()}>" for piece in all_segments)
  356 +
  357 +
  358 +def test_local_backend_batches_after_segmentation():
  359 + backend = _SegmentingMarianBackend(
  360 + name="opus-mt-en-zh",
  361 + model_id="Helsinki-NLP/opus-mt-en-zh",
  362 + model_dir="./models/translation/Helsinki-NLP/opus-mt-en-zh",
  363 + device="cpu",
  364 + torch_dtype="float32",
  365 + batch_size=4,
  366 + max_input_length=24,
  367 + max_new_tokens=24,
  368 + num_beams=1,
  369 + source_langs=["en"],
  370 + target_langs=["zh"],
  371 + )
  372 +
  373 + texts = [
  374 + "alpha beta gamma delta, epsilon zeta eta theta, iota kappa lambda mu.",
  375 + "nu xi omicron pi, rho sigma tau upsilon, phi chi psi omega.",
  376 + "dress shirt coat pants, socks shoes belt scarf, hat gloves bag watch.",
  377 + ]
  378 +
  379 + result = backend.translate(texts, source_lang="en", target_lang="zh")
  380 +
  381 + assert isinstance(result, list)
  382 + assert len(result) == 3
  383 + assert len(backend.translated_batches) >= 2
  384 + assert all(len(batch) <= backend.batch_size for batch in backend.translated_batches)
  385 + assert sum(len(batch) for batch in backend.translated_batches) > backend.batch_size
  386 + assert all(item is not None for item in result)
  387 +
  388 +
  389 +def test_local_backend_logs_segmentation_and_inference_batches(caplog):
  390 + backend = _SegmentingMarianBackend(
  391 + name="opus-mt-en-zh",
  392 + model_id="Helsinki-NLP/opus-mt-en-zh",
  393 + model_dir="./models/translation/Helsinki-NLP/opus-mt-en-zh",
  394 + device="cpu",
  395 + torch_dtype="float32",
  396 + batch_size=2,
  397 + max_input_length=24,
  398 + max_new_tokens=24,
  399 + num_beams=1,
  400 + source_langs=["en"],
  401 + target_langs=["zh"],
  402 + )
  403 +
  404 + texts = [
  405 + "one two three four, five six seven eight, nine ten eleven twelve.",
  406 + "thirteen fourteen fifteen sixteen, seventeen eighteen nineteen twenty.",
  407 + ]
  408 +
  409 + with caplog.at_level(logging.INFO):
  410 + backend.translate(texts, source_lang="en", target_lang="zh")
  411 +
  412 + messages = [record.getMessage() for record in caplog.records]
  413 +
  414 + assert any(message.startswith("Translation segmentation summary |") for message in messages)
  415 + inference_logs = [
  416 + message for message in messages if message.startswith("Translation inference batch |")
  417 + ]
  418 + assert len(inference_logs) >= 2
... ...
tests/test_translator_failure_semantics.py
  Show/Hide comments   View file @14e67b7
  1 +import logging
  2 +
1 3 from translation.cache import TranslationCache
  4 +from translation.logging_utils import (
  5 + TranslationRequestFilter,
  6 + bind_translation_request_id,
  7 + reset_translation_request_id,
  8 +)
2 9 from translation.service import TranslationService
3 10  
4 11  
... ... @@ -107,3 +114,80 @@ def test_service_caches_all_capabilities(monkeypatch):
107 114 ("opus-mt-zh-en", "en", "连衣裙", "opus-mt-zh-en:连衣裙"),
108 115 ("opus-mt-zh-en", "en", "衬衫", "opus-mt-zh-en:衬衫"),
109 116 ]
  117 +
  118 +
  119 +def test_translation_request_filter_injects_reqid():
  120 + reqid, token = bind_translation_request_id("req-test-1234567890")
  121 + try:
  122 + record = logging.LogRecord(
  123 + name="translation.service",
  124 + level=logging.INFO,
  125 + pathname=__file__,
  126 + lineno=1,
  127 + msg="hello",
  128 + args=(),
  129 + exc_info=None,
  130 + )
  131 + TranslationRequestFilter().filter(record)
  132 +
  133 + assert reqid == "req-test-1234567890"
  134 + assert record.reqid == "req-test-1234567890"
  135 + finally:
  136 + reset_translation_request_id(token)
  137 +
  138 +
  139 +def test_translation_route_log_focuses_on_routing_decision(monkeypatch, caplog):
  140 + monkeypatch.setattr(TranslationCache, "_init_redis_client", staticmethod(lambda: None))
  141 +
  142 + def _fake_create_backend(self, *, name, backend_type, cfg):
  143 + del self, backend_type, cfg
  144 +
  145 + class _Backend:
  146 + model = name
  147 +
  148 + @property
  149 + def supports_batch(self):
  150 + return True
  151 +
  152 + def translate(self, text, target_lang, source_lang=None, scene=None):
  153 + del target_lang, source_lang, scene
  154 + return text
  155 +
  156 + return _Backend()
  157 +
  158 + monkeypatch.setattr(TranslationService, "_create_backend", _fake_create_backend)
  159 + service = TranslationService(
  160 + {
  161 + "service_url": "http://127.0.0.1:6006",
  162 + "timeout_sec": 10.0,
  163 + "default_model": "llm",
  164 + "default_scene": "general",
  165 + "capabilities": {
  166 + "llm": {
  167 + "enabled": True,
  168 + "backend": "llm",
  169 + "model": "dummy-llm",
  170 + "base_url": "https://example.com",
  171 + "timeout_sec": 10.0,
  172 + "use_cache": True,
  173 + }
  174 + },
  175 + "cache": {
  176 + "ttl_seconds": 60,
  177 + "sliding_expiration": True,
  178 + },
  179 + }
  180 + )
  181 +
  182 + with caplog.at_level(logging.INFO):
  183 + service.translate("商品标题", target_lang="en", source_lang="zh", model="llm")
  184 +
  185 + route_messages = [
  186 + record.getMessage()
  187 + for record in caplog.records
  188 + if record.name == "translation.service" and record.getMessage().startswith("Translation route |")
  189 + ]
  190 +
  191 + assert route_messages == [
  192 + "Translation route | backend=llm request_type=single use_cache=True cache_available=False"
  193 + ]
... ...
translation/backends/local_ctranslate2.py
  Show/Hide comments   View file @14e67b7
... ... @@ -13,7 +13,12 @@ from typing import Dict, List, Optional, Sequence, Union
13 13  
14 14 from transformers import AutoTokenizer
15 15  
16   -from translation.languages import MARIAN_LANGUAGE_DIRECTIONS, NLLB_LANGUAGE_CODES
  16 +from translation.languages import (
  17 + MARIAN_LANGUAGE_DIRECTIONS,
  18 + build_nllb_language_catalog,
  19 + normalize_language_key,
  20 + resolve_nllb_language_code,
  21 +)
17 22 from translation.text_splitter import (
18 23 compute_safe_input_token_limit,
19 24 join_translated_segments,
... ... @@ -23,6 +28,17 @@ from translation.text_splitter import (
23 28 logger = logging.getLogger(__name__)
24 29  
25 30  
  31 +def _text_preview(text: Optional[str], limit: int = 32) -> str:
  32 + return str(text or "").replace("\n", "\\n")[:limit]
  33 +
  34 +
  35 +def _summarize_lengths(values: Sequence[int]) -> str:
  36 + if not values:
  37 + return "[]"
  38 + total = sum(values)
  39 + return f"min={min(values)} max={max(values)} avg={total / len(values):.1f}"
  40 +
  41 +
26 42 def _resolve_device(device: Optional[str]) -> str:
27 43 value = str(device or "auto").strip().lower()
28 44 if value not in {"auto", "cpu", "cuda"}:
... ... @@ -285,6 +301,17 @@ class LocalCTranslate2TranslationBackend:
285 301 source_tokens = self._encode_source_tokens(texts, source_lang, target_lang)
286 302 target_prefix = self._target_prefixes(len(source_tokens), source_lang, target_lang)
287 303 max_decoding_length = self._resolve_max_decoding_length(source_tokens)
  304 + logger.info(
  305 + "Translation model batch detail | model=%s segment_count=%s token_lengths=%s max_decoding_length=%s batch_type=%s beam_size=%s target_lang=%s source_lang=%s",
  306 + self.model,
  307 + len(source_tokens),
  308 + _summarize_lengths([len(tokens) for tokens in source_tokens]),
  309 + max_decoding_length,
  310 + self.ct2_batch_type,
  311 + self.num_beams,
  312 + target_lang,
  313 + source_lang or "auto",
  314 + )
288 315 results = self.translator.translate_batch(
289 316 source_tokens,
290 317 target_prefix=target_prefix,
... ... @@ -336,6 +363,59 @@ class LocalCTranslate2TranslationBackend:
336 363 ),
337 364 )
338 365  
  366 + def _log_segmentation_summary(
  367 + self,
  368 + *,
  369 + texts: Sequence[str],
  370 + segment_plans: Sequence[Sequence[str]],
  371 + target_lang: str,
  372 + source_lang: Optional[str],
  373 + ) -> None:
  374 + non_empty_count = sum(1 for text in texts if text.strip())
  375 + segment_counts = [len(segments) for segments in segment_plans if segments]
  376 + total_segments = sum(segment_counts)
  377 + segmented_inputs = sum(1 for count in segment_counts if count > 1)
  378 + logger.info(
  379 + "Translation segmentation summary | model=%s inputs=%s non_empty_inputs=%s segmented_inputs=%s total_segments=%s batch_size=%s target_lang=%s source_lang=%s segments_per_input=%s",
  380 + self.model,
  381 + len(texts),
  382 + non_empty_count,
  383 + segmented_inputs,
  384 + total_segments,
  385 + self.batch_size,
  386 + target_lang,
  387 + source_lang or "auto",
  388 + _summarize_lengths(segment_counts),
  389 + )
  390 +
  391 + def _translate_segment_batches(
  392 + self,
  393 + segments: List[str],
  394 + target_lang: str,
  395 + source_lang: Optional[str] = None,
  396 + ) -> List[Optional[str]]:
  397 + if not segments:
  398 + return []
  399 + outputs: List[Optional[str]] = []
  400 + total_batches = (len(segments) + self.batch_size - 1) // self.batch_size
  401 + for batch_index, start in enumerate(range(0, len(segments), self.batch_size), start=1):
  402 + batch = segments[start:start + self.batch_size]
  403 + logger.info(
  404 + "Translation inference batch | model=%s batch_index=%s total_batches=%s segment_count=%s char_lengths=%s first_preview=%s target_lang=%s source_lang=%s",
  405 + self.model,
  406 + batch_index,
  407 + total_batches,
  408 + len(batch),
  409 + _summarize_lengths([len(segment) for segment in batch]),
  410 + _text_preview(batch[0] if batch else ""),
  411 + target_lang,
  412 + source_lang or "auto",
  413 + )
  414 + outputs.extend(
  415 + self._translate_batch(batch, target_lang=target_lang, source_lang=source_lang)
  416 + )
  417 + return outputs
  418 +
339 419 def _translate_with_segmentation(
340 420 self,
341 421 texts: List[str],
... ... @@ -352,8 +432,15 @@ class LocalCTranslate2TranslationBackend:
352 432 segment_plans.append(segments)
353 433 flat_segments.extend(segments)
354 434  
  435 + self._log_segmentation_summary(
  436 + texts=texts,
  437 + segment_plans=segment_plans,
  438 + target_lang=target_lang,
  439 + source_lang=source_lang,
  440 + )
  441 +
355 442 translated_segments = (
356   - self._translate_batch(flat_segments, target_lang=target_lang, source_lang=source_lang)
  443 + self._translate_segment_batches(flat_segments, target_lang=target_lang, source_lang=source_lang)
357 444 if flat_segments
358 445 else []
359 446 )
... ... @@ -387,13 +474,10 @@ class LocalCTranslate2TranslationBackend:
387 474 del scene
388 475 is_single = isinstance(text, str)
389 476 texts = self._normalize_texts(text)
390   - outputs: List[Optional[str]] = []
391   - for start in range(0, len(texts), self.batch_size):
392   - chunk = texts[start:start + self.batch_size]
393   - if not any(item.strip() for item in chunk):
394   - outputs.extend([None if not item.strip() else item for item in chunk]) # type: ignore[list-item]
395   - continue
396   - outputs.extend(self._translate_with_segmentation(chunk, target_lang=target_lang, source_lang=source_lang))
  477 + if not any(item.strip() for item in texts):
  478 + outputs = [None if not item.strip() else item for item in texts] # type: ignore[list-item]
  479 + return outputs[0] if is_single else outputs
  480 + outputs = self._translate_with_segmentation(texts, target_lang=target_lang, source_lang=source_lang)
397 481 return outputs[0] if is_single else outputs
398 482  
399 483  
... ... @@ -492,11 +576,7 @@ class NLLBCTranslate2TranslationBackend(LocalCTranslate2TranslationBackend):
492 576 ct2_decoding_length_extra: int = 0,
493 577 ct2_decoding_length_min: int = 1,
494 578 ) -> None:
495   - overrides = language_codes or {}
496   - self.language_codes = {
497   - **NLLB_LANGUAGE_CODES,
498   - **{str(k).strip().lower(): str(v).strip() for k, v in overrides.items() if str(k).strip()},
499   - }
  579 + self.language_codes = build_nllb_language_catalog(language_codes)
500 580 self._tokenizers_by_source: Dict[str, object] = {}
501 581 super().__init__(
502 582 name=name,
... ... @@ -522,17 +602,17 @@ class NLLBCTranslate2TranslationBackend(LocalCTranslate2TranslationBackend):
522 602 )
523 603  
524 604 def _validate_languages(self, source_lang: Optional[str], target_lang: str) -> None:
525   - src = str(source_lang or "").strip().lower()
526   - tgt = str(target_lang or "").strip().lower()
527   - if not src:
  605 + if not str(source_lang or "").strip():
528 606 raise ValueError(f"Model '{self.model}' requires source_lang")
529   - if src not in self.language_codes:
  607 + if resolve_nllb_language_code(source_lang, self.language_codes) is None:
530 608 raise ValueError(f"Unsupported NLLB source language: {source_lang}")
531   - if tgt not in self.language_codes:
  609 + if resolve_nllb_language_code(target_lang, self.language_codes) is None:
532 610 raise ValueError(f"Unsupported NLLB target language: {target_lang}")
533 611  
534 612 def _get_tokenizer_for_source(self, source_lang: str):
535   - src_code = self.language_codes[source_lang]
  613 + src_code = resolve_nllb_language_code(source_lang, self.language_codes)
  614 + if src_code is None:
  615 + raise ValueError(f"Unsupported NLLB source language: {source_lang}")
536 616 with self._tokenizer_lock:
537 617 tokenizer = self._tokenizers_by_source.get(src_code)
538 618 if tokenizer is None:
... ... @@ -549,7 +629,7 @@ class NLLBCTranslate2TranslationBackend(LocalCTranslate2TranslationBackend):
549 629 target_lang: str,
550 630 ) -> List[List[str]]:
551 631 del target_lang
552   - source_key = str(source_lang or "").strip().lower()
  632 + source_key = normalize_language_key(source_lang)
553 633 tokenizer = self._get_tokenizer_for_source(source_key)
554 634 encoded = tokenizer(
555 635 texts,
... ... @@ -567,7 +647,9 @@ class NLLBCTranslate2TranslationBackend(LocalCTranslate2TranslationBackend):
567 647 target_lang: str,
568 648 ) -> Optional[List[Optional[List[str]]]]:
569 649 del source_lang
570   - tgt_code = self.language_codes[str(target_lang).strip().lower()]
  650 + tgt_code = resolve_nllb_language_code(target_lang, self.language_codes)
  651 + if tgt_code is None:
  652 + raise ValueError(f"Unsupported NLLB target language: {target_lang}")
571 653 return [[tgt_code] for _ in range(count)]
572 654  
573 655 def _postprocess_hypothesis(
... ... @@ -577,7 +659,9 @@ class NLLBCTranslate2TranslationBackend(LocalCTranslate2TranslationBackend):
577 659 target_lang: str,
578 660 ) -> List[str]:
579 661 del source_lang
580   - tgt_code = self.language_codes[str(target_lang).strip().lower()]
  662 + tgt_code = resolve_nllb_language_code(target_lang, self.language_codes)
  663 + if tgt_code is None:
  664 + raise ValueError(f"Unsupported NLLB target language: {target_lang}")
581 665 if tokens and tokens[0] == tgt_code:
582 666 return tokens[1:]
583 667 return tokens
... ...
translation/backends/local_seq2seq.py
  Show/Hide comments   View file @14e67b7
... ... @@ -10,7 +10,11 @@ from typing import Dict, List, Optional, Sequence, Union
10 10 import torch
11 11 from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
12 12  
13   -from translation.languages import MARIAN_LANGUAGE_DIRECTIONS, NLLB_LANGUAGE_CODES
  13 +from translation.languages import (
  14 + MARIAN_LANGUAGE_DIRECTIONS,
  15 + build_nllb_language_catalog,
  16 + resolve_nllb_language_code,
  17 +)
14 18 from translation.text_splitter import (
15 19 compute_safe_input_token_limit,
16 20 join_translated_segments,
... ... @@ -20,6 +24,17 @@ from translation.text_splitter import (
20 24 logger = logging.getLogger(__name__)
21 25  
22 26  
  27 +def _text_preview(text: Optional[str], limit: int = 32) -> str:
  28 + return str(text or "").replace("\n", "\\n")[:limit]
  29 +
  30 +
  31 +def _summarize_lengths(values: Sequence[int]) -> str:
  32 + if not values:
  33 + return "[]"
  34 + total = sum(values)
  35 + return f"min={min(values)} max={max(values)} avg={total / len(values):.1f}"
  36 +
  37 +
23 38 def _resolve_device(device: Optional[str]) -> str:
24 39 value = str(device or "auto").strip().lower()
25 40 if value == "auto":
... ... @@ -198,6 +213,59 @@ class LocalSeq2SeqTranslationBackend:
198 213 ),
199 214 )
200 215  
  216 + def _log_segmentation_summary(
  217 + self,
  218 + *,
  219 + texts: Sequence[str],
  220 + segment_plans: Sequence[Sequence[str]],
  221 + target_lang: str,
  222 + source_lang: Optional[str],
  223 + ) -> None:
  224 + non_empty_count = sum(1 for text in texts if text.strip())
  225 + segment_counts = [len(segments) for segments in segment_plans if segments]
  226 + total_segments = sum(segment_counts)
  227 + segmented_inputs = sum(1 for count in segment_counts if count > 1)
  228 + logger.info(
  229 + "Translation segmentation summary | model=%s inputs=%s non_empty_inputs=%s segmented_inputs=%s total_segments=%s batch_size=%s target_lang=%s source_lang=%s segments_per_input=%s",
  230 + self.model,
  231 + len(texts),
  232 + non_empty_count,
  233 + segmented_inputs,
  234 + total_segments,
  235 + self.batch_size,
  236 + target_lang,
  237 + source_lang or "auto",
  238 + _summarize_lengths(segment_counts),
  239 + )
  240 +
  241 + def _translate_segment_batches(
  242 + self,
  243 + segments: List[str],
  244 + target_lang: str,
  245 + source_lang: Optional[str] = None,
  246 + ) -> List[Optional[str]]:
  247 + if not segments:
  248 + return []
  249 + outputs: List[Optional[str]] = []
  250 + total_batches = (len(segments) + self.batch_size - 1) // self.batch_size
  251 + for batch_index, start in enumerate(range(0, len(segments), self.batch_size), start=1):
  252 + batch = segments[start:start + self.batch_size]
  253 + logger.info(
  254 + "Translation inference batch | model=%s batch_index=%s total_batches=%s segment_count=%s char_lengths=%s first_preview=%s target_lang=%s source_lang=%s",
  255 + self.model,
  256 + batch_index,
  257 + total_batches,
  258 + len(batch),
  259 + _summarize_lengths([len(segment) for segment in batch]),
  260 + _text_preview(batch[0] if batch else ""),
  261 + target_lang,
  262 + source_lang or "auto",
  263 + )
  264 + outputs.extend(
  265 + self._translate_batch(batch, target_lang=target_lang, source_lang=source_lang)
  266 + )
  267 + return outputs
  268 +
201 269 def _translate_with_segmentation(
202 270 self,
203 271 texts: List[str],
... ... @@ -214,8 +282,15 @@ class LocalSeq2SeqTranslationBackend:
214 282 segment_plans.append(segments)
215 283 flat_segments.extend(segments)
216 284  
  285 + self._log_segmentation_summary(
  286 + texts=texts,
  287 + segment_plans=segment_plans,
  288 + target_lang=target_lang,
  289 + source_lang=source_lang,
  290 + )
  291 +
217 292 translated_segments = (
218   - self._translate_batch(flat_segments, target_lang=target_lang, source_lang=source_lang)
  293 + self._translate_segment_batches(flat_segments, target_lang=target_lang, source_lang=source_lang)
219 294 if flat_segments
220 295 else []
221 296 )
... ... @@ -249,13 +324,10 @@ class LocalSeq2SeqTranslationBackend:
249 324 del scene
250 325 is_single = isinstance(text, str)
251 326 texts = self._normalize_texts(text)
252   - outputs: List[Optional[str]] = []
253   - for start in range(0, len(texts), self.batch_size):
254   - chunk = texts[start:start + self.batch_size]
255   - if not any(item.strip() for item in chunk):
256   - outputs.extend([None if not item.strip() else item for item in chunk]) # type: ignore[list-item]
257   - continue
258   - outputs.extend(self._translate_with_segmentation(chunk, target_lang=target_lang, source_lang=source_lang))
  327 + if not any(item.strip() for item in texts):
  328 + outputs = [None if not item.strip() else item for item in texts] # type: ignore[list-item]
  329 + return outputs[0] if is_single else outputs
  330 + outputs = self._translate_with_segmentation(texts, target_lang=target_lang, source_lang=source_lang)
259 331 return outputs[0] if is_single else outputs
260 332  
261 333  
... ... @@ -324,11 +396,7 @@ class NLLBTranslationBackend(LocalSeq2SeqTranslationBackend):
324 396 language_codes: Optional[Dict[str, str]] = None,
325 397 attn_implementation: Optional[str] = None,
326 398 ) -> None:
327   - overrides = language_codes or {}
328   - self.language_codes = {
329   - **NLLB_LANGUAGE_CODES,
330   - **{str(k).strip().lower(): str(v).strip() for k, v in overrides.items() if str(k).strip()},
331   - }
  399 + self.language_codes = build_nllb_language_catalog(language_codes)
332 400 super().__init__(
333 401 name=name,
334 402 model_id=model_id,
... ... @@ -343,24 +411,26 @@ class NLLBTranslationBackend(LocalSeq2SeqTranslationBackend):
343 411 )
344 412  
345 413 def _validate_languages(self, source_lang: Optional[str], target_lang: str) -> None:
346   - src = str(source_lang or "").strip().lower()
347   - tgt = str(target_lang or "").strip().lower()
348   - if not src:
  414 + if not str(source_lang or "").strip():
349 415 raise ValueError(f"Model '{self.model}' requires source_lang")
350   - if src not in self.language_codes:
  416 + if resolve_nllb_language_code(source_lang, self.language_codes) is None:
351 417 raise ValueError(f"Unsupported NLLB source language: {source_lang}")
352   - if tgt not in self.language_codes:
  418 + if resolve_nllb_language_code(target_lang, self.language_codes) is None:
353 419 raise ValueError(f"Unsupported NLLB target language: {target_lang}")
354 420  
355 421 def _prepare_tokenizer(self, source_lang: Optional[str], target_lang: str) -> Dict[str, object]:
356 422 del target_lang
357   - src_code = self.language_codes[str(source_lang).strip().lower()]
  423 + src_code = resolve_nllb_language_code(source_lang, self.language_codes)
  424 + if src_code is None:
  425 + raise ValueError(f"Unsupported NLLB source language: {source_lang}")
358 426 self.tokenizer.src_lang = src_code
359 427 return {}
360 428  
361 429 def _build_generate_kwargs(self, source_lang: Optional[str], target_lang: str) -> Dict[str, object]:
362 430 del source_lang
363   - tgt_code = self.language_codes[str(target_lang).strip().lower()]
  431 + tgt_code = resolve_nllb_language_code(target_lang, self.language_codes)
  432 + if tgt_code is None:
  433 + raise ValueError(f"Unsupported NLLB target language: {target_lang}")
364 434 forced_bos_token_id = None
365 435 if hasattr(self.tokenizer, "lang_code_to_id"):
366 436 forced_bos_token_id = self.tokenizer.lang_code_to_id.get(tgt_code)
... ...
translation/cache.py
  Show/Hide comments   View file @14e67b7
... ... @@ -40,10 +40,8 @@ class TranslationCache:
40 40 try:
41 41 value = self.redis_client.get(key)
42 42 logger.info(
43   - "Translation cache %s | model=%s target_lang=%s text_len=%s key=%s",
  43 + "Translation cache %s | text_len=%s key=%s",
44 44 "hit" if value is not None else "miss",
45   - model,
46   - target_lang,
47 45 len(str(source_text or "")),
48 46 key,
49 47 )
... ... @@ -61,9 +59,7 @@ class TranslationCache:
61 59 try:
62 60 self.redis_client.setex(key, self.ttl_seconds, translated_text)
63 61 logger.info(
64   - "Translation cache write | model=%s target_lang=%s text_len=%s result_len=%s ttl_seconds=%s key=%s",
65   - model,
66   - target_lang,
  62 + "Translation cache write | text_len=%s result_len=%s ttl_seconds=%s key=%s",
67 63 len(str(source_text or "")),
68 64 len(str(translated_text or "")),
69 65 self.ttl_seconds,
... ...
translation/languages.py
  Show/Hide comments   View file @14e67b7
... ... @@ -2,12 +2,13 @@
2 2  
3 3 from __future__ import annotations
4 4  
5   -from typing import Dict, Tuple
  5 +from typing import Dict, Mapping, Optional, Tuple
6 6  
7 7  
8 8 LANGUAGE_LABELS: Dict[str, str] = {
9 9 "zh": "Chinese",
10 10 "en": "English",
  11 + "fi": "Finnish",
11 12 "ru": "Russian",
12 13 "ar": "Arabic",
13 14 "ja": "Japanese",
... ... @@ -49,6 +50,7 @@ DEEPL_LANGUAGE_CODES: Dict[str, str] = {
49 50  
50 51 NLLB_LANGUAGE_CODES: Dict[str, str] = {
51 52 "en": "eng_Latn",
  53 + "fi": "fin_Latn",
52 54 "zh": "zho_Hans",
53 55 "ru": "rus_Cyrl",
54 56 "ar": "arb_Arab",
... ... @@ -65,3 +67,56 @@ MARIAN_LANGUAGE_DIRECTIONS: Dict[str, Tuple[str, str]] = {
65 67 "opus-mt-zh-en": ("zh", "en"),
66 68 "opus-mt-en-zh": ("en", "zh"),
67 69 }
  70 +
  71 +
  72 +NLLB_LANGUAGE_ALIASES: Dict[str, str] = {
  73 + "fi_fi": "fi",
  74 + "fin": "fi",
  75 + "fin_fin": "fi",
  76 + "zh_cn": "zh",
  77 + "zh_hans": "zh",
  78 +}
  79 +
  80 +
  81 +def normalize_language_key(language: Optional[str]) -> str:
  82 + return str(language or "").strip().lower().replace("-", "_")
  83 +
  84 +
  85 +def build_nllb_language_catalog(
  86 + overrides: Optional[Mapping[str, str]] = None,
  87 +) -> Dict[str, str]:
  88 + catalog = {
  89 + normalize_language_key(key): str(value).strip()
  90 + for key, value in NLLB_LANGUAGE_CODES.items()
  91 + if str(key).strip()
  92 + }
  93 + for key, value in (overrides or {}).items():
  94 + normalized_key = normalize_language_key(key)
  95 + if normalized_key:
  96 + catalog[normalized_key] = str(value).strip()
  97 + return catalog
  98 +
  99 +
  100 +def resolve_nllb_language_code(
  101 + language: Optional[str],
  102 + language_codes: Optional[Mapping[str, str]] = None,
  103 +) -> Optional[str]:
  104 + normalized = normalize_language_key(language)
  105 + if not normalized:
  106 + return None
  107 +
  108 + catalog = build_nllb_language_catalog(language_codes)
  109 + direct = catalog.get(normalized)
  110 + if direct is not None:
  111 + return direct
  112 +
  113 + alias = NLLB_LANGUAGE_ALIASES.get(normalized)
  114 + if alias is not None:
  115 + aliased = catalog.get(normalize_language_key(alias))
  116 + if aliased is not None:
  117 + return aliased
  118 +
  119 + for code in catalog.values():
  120 + if normalize_language_key(code) == normalized:
  121 + return code
  122 + return None
... ...
translation/logging_utils.py 0 → 100644
  Show/Hide comments   View file @14e67b7
... ... @@ -0,0 +1,37 @@
  1 +"""Shared translation logging context helpers."""
  2 +
  3 +from __future__ import annotations
  4 +
  5 +import contextvars
  6 +import logging
  7 +import uuid
  8 +from typing import Optional
  9 +
  10 +
  11 +_translation_request_id_var: contextvars.ContextVar[Optional[str]] = contextvars.ContextVar(
  12 + "translation_request_id",
  13 + default=None,
  14 +)
  15 +
  16 +
  17 +def current_translation_request_id() -> str:
  18 + return _translation_request_id_var.get() or "-1"
  19 +
  20 +
  21 +def bind_translation_request_id(request_id: Optional[str] = None) -> tuple[str, contextvars.Token]:
  22 + raw_value = str(request_id or "").strip()
  23 + normalized = raw_value[:32] if raw_value else str(uuid.uuid4())[:8]
  24 + return normalized, _translation_request_id_var.set(normalized)
  25 +
  26 +
  27 +def reset_translation_request_id(token: contextvars.Token) -> None:
  28 + _translation_request_id_var.reset(token)
  29 +
  30 +
  31 +class TranslationRequestFilter(logging.Filter):
  32 + """Inject a request id into translator logs when one is available."""
  33 +
  34 + def filter(self, record: logging.LogRecord) -> bool:
  35 + if not hasattr(record, "reqid"):
  36 + record.reqid = current_translation_request_id()
  37 + return True
... ...
translation/service.py
  Show/Hide comments   View file @14e67b7
... ... @@ -198,15 +198,10 @@ class TranslationService:
198 198 active_scene = normalize_translation_scene(self.config, scene)
199 199 capability_cfg = self._enabled_capabilities[normalized_model]
200 200 use_cache = bool(capability_cfg.get("use_cache"))
201   - text_count = 1 if isinstance(text, str) else len(list(text))
202 201 logger.info(
203   - "Translation route | model=%s backend=%s scene=%s target_lang=%s source_lang=%s count=%s use_cache=%s cache_available=%s",
204   - normalized_model,
  202 + "Translation route | backend=%s request_type=%s use_cache=%s cache_available=%s",
205 203 getattr(backend, "model", normalized_model),
206   - active_scene,
207   - target_lang,
208   - source_lang or "auto",
209   - text_count,
  204 + "single" if isinstance(text, str) else "batch",
210 205 use_cache,
211 206 self._translation_cache.available,
212 207 )
... ... @@ -252,11 +247,7 @@ class TranslationService:
252 247 cached = self._translation_cache.get(model=model, target_lang=target_lang, source_text=text)
253 248 if cached is not None:
254 249 logger.info(
255   - "Translation cache served | model=%s scene=%s target_lang=%s source_lang=%s text_len=%s",
256   - model,
257   - scene,
258   - target_lang,
259   - source_lang or "auto",
  250 + "Translation cache served | request_type=single text_len=%s",
260 251 len(text),
261 252 )
262 253 return cached
... ... @@ -274,21 +265,13 @@ class TranslationService:
274 265 translated_text=translated,
275 266 )
276 267 logger.info(
277   - "Translation backend result cached | model=%s scene=%s target_lang=%s source_lang=%s text_len=%s result_len=%s",
278   - model,
279   - scene,
280   - target_lang,
281   - source_lang or "auto",
  268 + "Translation backend result cached | request_type=single text_len=%s result_len=%s",
282 269 len(text),
283 270 len(str(translated)),
284 271 )
285 272 else:
286 273 logger.warning(
287   - "Translation backend returned empty result | model=%s scene=%s target_lang=%s source_lang=%s text_len=%s",
288   - model,
289   - scene,
290   - target_lang,
291   - source_lang or "auto",
  274 + "Translation backend returned empty result | request_type=single text_len=%s",
292 275 len(text),
293 276 )
294 277 return translated
... ... @@ -327,11 +310,7 @@ class TranslationService:
327 310 miss_indices.append(idx)
328 311  
329 312 logger.info(
330   - "Translation batch cache summary | model=%s scene=%s target_lang=%s source_lang=%s total=%s cache_hits=%s cache_misses=%s",
331   - model,
332   - scene,
333   - target_lang,
334   - source_lang or "auto",
  313 + "Translation batch cache summary | total=%s cache_hits=%s cache_misses=%s",
335 314 len(texts),
336 315 cache_hits,
337 316 len(misses),
... ... @@ -356,11 +335,7 @@ class TranslationService:
356 335 )
357 336 else:
358 337 logger.warning(
359   - "Translation batch item returned empty result | model=%s scene=%s target_lang=%s source_lang=%s item_index=%s text_len=%s",
360   - model,
361   - scene,
362   - target_lang,
363   - source_lang or "auto",
  338 + "Translation batch item returned empty result | item_index=%s text_len=%s",
364 339 idx,
365 340 len(original_text),
366 341 )
... ...