mirror of
https://github.com/OpenByteInc/QuantDinger.git
synced 2026-09-28 23:32:55 +08:00
274 lines
9.5 KiB
Python
274 lines
9.5 KiB
Python
"""
|
||
Indicator Translator - 指标多语言翻译服务
|
||
|
||
设计目标
|
||
========
|
||
解决「指标市场里来自全球作者的 name/description 语言混杂、用户看不懂」的问题。
|
||
做法：作者点发布时，调用 LLM 把 name + description 同时翻译成全部支持语言，
|
||
将结果写到 qd_indicator_codes.name_i18n / description_i18n (JSONB 字典)，
|
||
读取接口按 Accept-Language 取对应键即可。
|
||
|
||
支持的语言列表来源
|
||
==================
|
||
QuantDinger-Vue-src/src/locales/lang/ 下的所有 .js — 10 种：
|
||
zh-CN, en-US, zh-TW, ar-SA, de-DE, fr-FR, ja-JP, ko-KR, th-TH, vi-VN。
|
||
|
||
容错策略
|
||
========
|
||
LLM 调用本身是网络密集且偶尔会失败的：
|
||
1. 任何步骤抛异常时，本模块永远 *不会* 让上游 save_indicator 失败 ——
|
||
只在日志里 warn，并把 name_i18n/description_i18n 留为 None；
|
||
下游接口会自动 fallback 到原始 name/description。
|
||
2. 如果 LLM 返回的 JSON 缺某些语言，沿用已有 / 留空（仍可 fallback）。
|
||
3. 原始语言对应键直接写入用户原文，不再过 LLM，避免 LLM 译回原文丢失味道。
|
||
4. 翻译是 "best-effort"：即使所有语言都翻译失败，只要 source_language 设置正确，
|
||
接口仍然能展示原文（只是非原生用户体验下降，不会 5xx）。
|
||
|
||
性能考虑
|
||
========
|
||
- 同步阻塞：save_indicator 等待翻译完成再返回。一次 LLM 调用约 2-5s，
|
||
对作者发布操作来说在可接受范围。如果以后 P99 拖太长，可改成后台 worker。
|
||
- 单次调用：一次 prompt 让 LLM 输出所有目标语言的 JSON，比 10 次单独调用快得多。
|
||
"""
|
||
from __future__ import annotations
|
||
|
||
import json
|
||
import re
|
||
from typing import Dict, Optional, Tuple
|
||
|
||
from app.services.llm import LLMService
|
||
from app.utils.logger import get_logger
|
||
|
||
logger = get_logger(__name__)
|
||
|
||
|
||
SUPPORTED_LANGUAGES: Dict[str, str] = {
|
||
'zh-CN': 'Simplified Chinese',
|
||
'zh-TW': 'Traditional Chinese',
|
||
'en-US': 'English',
|
||
'ar-SA': 'Arabic',
|
||
'de-DE': 'German',
|
||
'fr-FR': 'French',
|
||
'ja-JP': 'Japanese',
|
||
'ko-KR': 'Korean',
|
||
'th-TH': 'Thai',
|
||
'vi-VN': 'Vietnamese',
|
||
}
|
||
|
||
|
||
_ASCII_RATIO_FOR_LATIN = 0.85
|
||
|
||
|
||
def detect_source_language(text: str, fallback: str = 'en-US') -> str:
|
||
"""根据文本字符分布做一个粗糙的语言检测，仅用于兜底。
|
||
|
||
精确语言检测建议在前端按用户 UI 语言决定后传给后端，本函数只在前端没传时
|
||
避免空值。常见场景：
|
||
- 中文比例 > 30% → zh-CN
|
||
- 日文假名出现 → ja-JP
|
||
- 韩文音节出现 → ko-KR
|
||
- 阿拉伯字符出现 → ar-SA
|
||
- 泰文字符出现 → th-TH
|
||
- 越南组合元音 → vi-VN
|
||
- 其它 → fallback (默认 en-US)
|
||
"""
|
||
if not text:
|
||
return fallback
|
||
|
||
has_cjk = bool(re.search(r'[\u4e00-\u9fff]', text))
|
||
has_kana = bool(re.search(r'[\u3040-\u30ff]', text))
|
||
has_hangul = bool(re.search(r'[\uac00-\ud7af]', text))
|
||
has_arabic = bool(re.search(r'[\u0600-\u06ff]', text))
|
||
has_thai = bool(re.search(r'[\u0e00-\u0e7f]', text))
|
||
has_viet = bool(re.search(r'[\u00C0-\u1EF9]', text))
|
||
|
||
if has_kana: return 'ja-JP'
|
||
if has_hangul: return 'ko-KR'
|
||
if has_arabic: return 'ar-SA'
|
||
if has_thai: return 'th-TH'
|
||
if has_cjk: return 'zh-CN'
|
||
if has_viet: return 'vi-VN'
|
||
return fallback
|
||
|
||
|
||
def _build_prompt(name: str, description: str, source_lang: str) -> Tuple[str, str]:
|
||
"""组装 system + user prompt。"""
|
||
target_pairs = "\n".join(
|
||
f" - {code}: {label}"
|
||
for code, label in SUPPORTED_LANGUAGES.items()
|
||
)
|
||
|
||
system_prompt = (
|
||
"You are a professional translator specialised in quantitative trading and "
|
||
"technical indicators. Translate the given indicator name and description into "
|
||
"all of the following languages.\n\n"
|
||
"Rules:\n"
|
||
"1. Preserve all technical jargon (RSI, MACD, EMA, Bollinger Bands, "
|
||
"ATR, ADX, etc.) verbatim in every language — never translate the "
|
||
"ticker / formula names.\n"
|
||
"2. Keep the name SHORT (ideally <= 4 words / <=14 CJK characters).\n"
|
||
"3. Keep the description tight: 1-2 sentences, plain prose, no markdown, "
|
||
"no emojis, no quotes.\n"
|
||
"4. If a target language is the same as the source language, copy the "
|
||
"original text unchanged.\n"
|
||
"5. Output STRICT JSON only — no markdown fences, no commentary.\n\n"
|
||
f"Target languages:\n{target_pairs}\n\n"
|
||
'Output schema (every code must appear, even if value is identical to source):\n'
|
||
'{\n'
|
||
' "name": { "<code>": "...", ... },\n'
|
||
' "description": { "<code>": "...", ... }\n'
|
||
'}'
|
||
)
|
||
|
||
user_prompt = (
|
||
f"Source language: {source_lang} ({SUPPORTED_LANGUAGES.get(source_lang, source_lang)})\n\n"
|
||
f"Indicator name:\n{name}\n\n"
|
||
f"Indicator description:\n{description}\n"
|
||
)
|
||
|
||
return system_prompt, user_prompt
|
||
|
||
|
||
def _coerce_str(val) -> str:
|
||
"""LLM 偶尔输出 list / dict / None — 兜底转 str。"""
|
||
if val is None:
|
||
return ''
|
||
if isinstance(val, str):
|
||
return val.strip()
|
||
try:
|
||
return str(val).strip()
|
||
except Exception:
|
||
return ''
|
||
|
||
|
||
def translate_indicator(
|
||
name: str,
|
||
description: str,
|
||
source_language: Optional[str] = None,
|
||
) -> Tuple[Optional[Dict[str, str]], Optional[Dict[str, str]], str]:
|
||
"""把指标 name / description 翻译成全部支持的语言。
|
||
|
||
Args:
|
||
name: 原始 name (允许任意语言)
|
||
description: 原始 description
|
||
source_language: 原始语言代码 (zh-CN / en-US / ...)。若为 None 或不在支持列表，
|
||
自动用 ``detect_source_language`` 做粗判。
|
||
|
||
Returns:
|
||
(name_i18n, description_i18n, resolved_source_language)
|
||
- name_i18n / description_i18n: dict[lang_code -> translated_text]
|
||
翻译失败时返回 None（save_indicator 视作 NULL 入库，读取接口走 fallback）。
|
||
- resolved_source_language: 最终确定的原始语言代码（已做兜底）。
|
||
"""
|
||
src = source_language if source_language in SUPPORTED_LANGUAGES else None
|
||
if not src:
|
||
src = detect_source_language(description or name, fallback='en-US')
|
||
|
||
name = (name or '').strip()
|
||
description = (description or '').strip()
|
||
|
||
if not name and not description:
|
||
return None, None, src
|
||
|
||
try:
|
||
system_prompt, user_prompt = _build_prompt(name, description, src)
|
||
|
||
llm = LLMService()
|
||
result = llm.safe_call_llm(
|
||
system_prompt=system_prompt,
|
||
user_prompt=user_prompt,
|
||
default_structure={'name': {}, 'description': {}},
|
||
)
|
||
|
||
raw_name = result.get('name') if isinstance(result, dict) else None
|
||
raw_desc = result.get('description') if isinstance(result, dict) else None
|
||
|
||
if not isinstance(raw_name, dict) and not isinstance(raw_desc, dict):
|
||
logger.warning(
|
||
"translate_indicator: LLM returned no usable dict (raw=%s)",
|
||
str(result)[:200],
|
||
)
|
||
return None, None, src
|
||
|
||
name_i18n: Dict[str, str] = {}
|
||
desc_i18n: Dict[str, str] = {}
|
||
|
||
for code in SUPPORTED_LANGUAGES.keys():
|
||
n_val = _coerce_str((raw_name or {}).get(code))
|
||
d_val = _coerce_str((raw_desc or {}).get(code))
|
||
|
||
if code == src:
|
||
if name: n_val = name
|
||
if description: d_val = description
|
||
|
||
if n_val:
|
||
name_i18n[code] = n_val
|
||
if d_val:
|
||
desc_i18n[code] = d_val
|
||
|
||
if src not in name_i18n and src not in desc_i18n:
|
||
logger.warning(
|
||
"translate_indicator: no entry for source language %s after coercion",
|
||
src,
|
||
)
|
||
return None, None, src
|
||
|
||
return (name_i18n or None), (desc_i18n or None), src
|
||
|
||
except Exception as e:
|
||
logger.warning(f"translate_indicator failed (non-fatal): {e}")
|
||
return None, None, src
|
||
|
||
|
||
def pick_localized(
|
||
raw_text: str,
|
||
i18n_payload,
|
||
accept_lang: str,
|
||
source_lang: Optional[str] = None,
|
||
) -> str:
|
||
"""根据请求语言选最合适的字段。
|
||
|
||
优先级：
|
||
1. accept_lang 精确匹配 i18n_payload
|
||
2. accept_lang 的主语言（zh-* → 任意 zh-*）模糊匹配
|
||
3. en-US (universal fallback)
|
||
4. source_lang 原始语言
|
||
5. 原始 raw_text
|
||
|
||
Args:
|
||
raw_text: indicators 表里原始 name 或 description 字段
|
||
i18n_payload: name_i18n 或 description_i18n (dict 或 JSON str 或 None)
|
||
accept_lang: 当前请求语言代码，例如 'zh-CN'
|
||
source_lang: 原始语言代码
|
||
"""
|
||
if not i18n_payload:
|
||
return raw_text or ''
|
||
|
||
if isinstance(i18n_payload, str):
|
||
try:
|
||
i18n_payload = json.loads(i18n_payload)
|
||
except Exception:
|
||
return raw_text or ''
|
||
|
||
if not isinstance(i18n_payload, dict):
|
||
return raw_text or ''
|
||
|
||
val = i18n_payload.get(accept_lang)
|
||
if val:
|
||
return val
|
||
|
||
if accept_lang and '-' in accept_lang:
|
||
prefix = accept_lang.split('-')[0].lower() + '-'
|
||
for k, v in i18n_payload.items():
|
||
if isinstance(k, str) and k.lower().startswith(prefix) and v:
|
||
return v
|
||
|
||
# English universal fallback
|
||
if 'en-US' in i18n_payload and i18n_payload['en-US']:
|
||
return i18n_payload['en-US']
|
||
|
||
if source_lang and source_lang in i18n_payload and i18n_payload[source_lang]:
|
||
return i18n_payload[source_lang]
|
||
|
||
return raw_text or ''
|