5e4dc8e4
 tangwang
翻译架构按“一个翻译服务 +
|
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
|
"""LLM-based translation backend."""
from __future__ import annotations
import logging
import os
import time
from typing import List, Optional, Sequence, Union
from openai import OpenAI
from config.env_config import DASHSCOPE_API_KEY
from config.services_config import get_translation_config
from config.translate_prompts import TRANSLATION_PROMPTS
from config.tenant_config_loader import SOURCE_LANG_CODE_MAP
logger = logging.getLogger(__name__)
DEFAULT_QWEN_BASE_URL = "https://dashscope-us.aliyuncs.com/compatible-mode/v1"
DEFAULT_LLM_MODEL = "qwen-flash"
def _build_prompt(
text: str,
*,
source_lang: Optional[str],
target_lang: str,
scene: Optional[str],
) -> str:
tgt = (target_lang or "").lower() or "en"
src = (source_lang or "auto").lower()
normalized_scene = (scene or "").strip() or "general"
if normalized_scene in {"query", "ecommerce_search", "ecommerce_search_query"}:
group_key = "ecommerce_search_query"
elif normalized_scene in {"product_title", "sku_name"}:
group_key = "sku_name"
else:
group_key = normalized_scene
group = TRANSLATION_PROMPTS.get(group_key) or TRANSLATION_PROMPTS["general"]
template = group.get(tgt) or group.get("en")
if not template:
template = (
"You are a professional {source_lang} ({src_lang_code}) to "
"{target_lang} ({tgt_lang_code}) translator, output only the translation: {text}"
)
source_lang_label = SOURCE_LANG_CODE_MAP.get(src, src)
target_lang_label = SOURCE_LANG_CODE_MAP.get(tgt, tgt)
return template.format(
source_lang=source_lang_label,
src_lang_code=src,
target_lang=target_lang_label,
tgt_lang_code=tgt,
text=text,
)
class LLMTranslationBackend:
def __init__(
self,
*,
model: Optional[str] = None,
timeout_sec: float = 30.0,
base_url: Optional[str] = None,
) -> None:
cfg = get_translation_config()
llm_cfg = cfg.get_capability_cfg("llm")
self.model = model or llm_cfg.get("model") or DEFAULT_LLM_MODEL
self.timeout_sec = float(llm_cfg.get("timeout_sec") or timeout_sec or 30.0)
self.base_url = (
(base_url or "").strip()
or (llm_cfg.get("base_url") or "").strip()
or os.getenv("DASHSCOPE_BASE_URL")
or DEFAULT_QWEN_BASE_URL
)
self.client = self._create_client()
@property
def supports_batch(self) -> bool:
return True
def _create_client(self) -> Optional[OpenAI]:
api_key = DASHSCOPE_API_KEY or os.getenv("DASHSCOPE_API_KEY")
if not api_key:
logger.warning("DASHSCOPE_API_KEY not set; llm translation unavailable")
return None
try:
return OpenAI(api_key=api_key, base_url=self.base_url)
except Exception as exc:
logger.error("Failed to initialize llm translation client: %s", exc, exc_info=True)
return None
def _translate_single(
self,
text: str,
target_lang: str,
source_lang: Optional[str] = None,
context: Optional[str] = None,
prompt: Optional[str] = None,
) -> Optional[str]:
if not text or not str(text).strip():
return text
if not self.client:
return None
tgt = (target_lang or "").lower() or "en"
src = (source_lang or "auto").lower()
scene = context or "default"
user_prompt = prompt or _build_prompt(
text=text,
source_lang=src,
target_lang=tgt,
scene=scene,
)
start = time.time()
try:
logger.info(
"[llm] Request | src=%s tgt=%s model=%s prompt=%s",
src,
tgt,
self.model,
user_prompt,
)
completion = self.client.chat.completions.create(
model=self.model,
messages=[{"role": "user", "content": user_prompt}],
timeout=self.timeout_sec,
)
content = (completion.choices[0].message.content or "").strip()
latency_ms = (time.time() - start) * 1000
if not content:
logger.warning("[llm] Empty result | src=%s tgt=%s latency=%.1fms", src, tgt, latency_ms)
return None
logger.info(
"[llm] Success | src=%s tgt=%s src_text=%s response=%s latency=%.1fms",
src,
tgt,
text,
content,
latency_ms,
)
return content
except Exception as exc:
latency_ms = (time.time() - start) * 1000
logger.warning(
"[llm] Failed | src=%s tgt=%s latency=%.1fms error=%s",
src,
tgt,
latency_ms,
exc,
exc_info=True,
)
return None
def translate(
self,
text: Union[str, Sequence[str]],
target_lang: str,
source_lang: Optional[str] = None,
context: Optional[str] = None,
prompt: Optional[str] = None,
) -> Union[Optional[str], List[Optional[str]]]:
if isinstance(text, (list, tuple)):
results: List[Optional[str]] = []
for item in text:
if item is None:
results.append(None)
continue
results.append(
self._translate_single(
text=str(item),
target_lang=target_lang,
source_lang=source_lang,
context=context,
prompt=prompt,
)
)
return results
return self._translate_single(
text=str(text),
target_lang=target_lang,
source_lang=source_lang,
context=context,
prompt=prompt,
)
LLMTranslatorProvider = LLMTranslationBackend
def llm_translate(
text: Union[str, Sequence[str]],
target_lang: str,
*,
source_lang: Optional[str] = None,
source_lang_label: Optional[str] = None,
target_lang_label: Optional[str] = None,
timeout_sec: Optional[float] = None,
) -> Union[Optional[str], List[Optional[str]]]:
del source_lang_label, target_lang_label
provider = LLMTranslationBackend(timeout_sec=timeout_sec or 30.0)
return provider.translate(
text=text,
target_lang=target_lang,
source_lang=source_lang,
context=None,
)
|