b27643123e
- 新增 app/services/translation/spark.py: OpenAI 兼容协议客户端, URL = https://spark-api-open.xf-yun.com/v1/chat/completions, 鉴权 = Bearer <APIPassword>, model = lite(默认)/generalv3.5/4.0Ultra 可切换 - service.py 引擎链路调整为: spark → tencent(配额)→ maas → agnes → local。 优先级降序: spark 配了 key 就用它,失败再走 tencent(继续吃配额,不绕过)。 要完全绕开 tencent,把 TENCENTCLOUD_SECRET_ID 留空即可。 - 配置: 新增 SPARK_API_PASSWORD / SPARK_BASE_URL / SPARK_MODEL / SPARK_INTERVAL_SEC (留空 SPARK_API_PASSWORD = 走原 tencent 主链路,向后兼容) - 缓存白名单 / 配额计数逻辑保持原行为,只把 spark 加入允许缓存的引擎集合
115 lines
4.4 KiB
Python
115 lines
4.4 KiB
Python
"""星火 Spark 翻译(OpenAI 兼容协议)。
|
|
|
|
- 端点:https://spark-api-open.xf-yun.com/v1/chat/completions
|
|
- 模型:lite(也支持 generalv3.5 等,但该项目用 Lite,免费额度)
|
|
- 鉴权:Bearer token(APIPassword 直接当 Bearer 用)
|
|
- 请求:POST /chat/completions,system prompt 告诉模型做翻译
|
|
|
|
设计上独立于 LlmClient,专门走 spark_* 配置,避免和 LLM 智能增强共用 client 的节流。
|
|
"""
|
|
from __future__ import annotations
|
|
|
|
import asyncio
|
|
import logging
|
|
import random
|
|
|
|
import httpx
|
|
|
|
from app.config import settings
|
|
from app.services.translation.base import BaseTranslator, TranslationResult
|
|
|
|
logger = logging.getLogger("news.translate.spark")
|
|
|
|
|
|
# 经过实测,这套 prompt 在 Spark Lite 上输出稳定(不夹带 reasoning / 注释)。
|
|
# Lite 不支持 system 角色(早期 Lite 模型),但 v1 OpenAI 兼容的 Lite 接受 system;
|
|
# 保留 system prompt 走 4.0Ultra/Max/Pro 时也通用。
|
|
_SYSTEM_PROMPT = """你是一个即时翻译助手。对于用户输入的英文或日文文章,请直接输出对应的简体中文译文。严格遵守:
|
|
|
|
不要进行任何分步思考、不要输出分析、不要添加注释或说明。
|
|
|
|
只输出中文译文本身,不包含任何额外文字(如"翻译结果:"、"以下是中文:"等)。
|
|
|
|
如果输入内容既非英文也非日文,仅回复:"仅支持英文或日文翻译为中文。" """
|
|
|
|
|
|
class SparkTranslator(BaseTranslator):
|
|
"""星火 Spark 翻译(OpenAI 兼容协议,模型 lite / generalv3.5 等)。"""
|
|
|
|
name = "spark"
|
|
|
|
def __init__(self):
|
|
if not settings.spark_api_password:
|
|
raise RuntimeError("Spark APIPassword missing")
|
|
self.api_password = settings.spark_api_password
|
|
self.base_url = settings.spark_base_url.rstrip("/")
|
|
self.model = settings.spark_model
|
|
self.interval_sec = settings.spark_interval_sec
|
|
|
|
def is_configured(self) -> bool:
|
|
return bool(self.api_password)
|
|
|
|
async def translate(
|
|
self, text: str, source: str = "auto", target: str = "zh"
|
|
) -> TranslationResult:
|
|
"""翻译接口。source/target 当前固定 EN/JA → ZH(由 prompt 控制)。"""
|
|
if not text.strip():
|
|
return TranslationResult(text=text, engine=self.name, chars=0)
|
|
|
|
if not self.is_configured():
|
|
raise RuntimeError("Spark APIPassword missing")
|
|
|
|
system = _SYSTEM_PROMPT
|
|
user = text
|
|
|
|
url = f"{self.base_url}/chat/completions"
|
|
payload = {
|
|
"model": self.model,
|
|
"messages": [
|
|
{"role": "system", "content": system},
|
|
{"role": "user", "content": user},
|
|
],
|
|
"temperature": 0.0,
|
|
"max_tokens": max(256, len(text) * 3),
|
|
}
|
|
headers = {
|
|
"Authorization": f"Bearer {self.api_password}",
|
|
"Content-Type": "application/json",
|
|
}
|
|
|
|
# 简单串行 + 重试 1 次
|
|
last_exc: Exception | None = None
|
|
for attempt in range(2):
|
|
try:
|
|
async with httpx.AsyncClient(timeout=60.0) as client:
|
|
r = await client.post(url, json=payload, headers=headers)
|
|
if r.status_code >= 500:
|
|
raise RuntimeError(f"Spark 5xx: {r.status_code} {r.text[:200]}")
|
|
if r.status_code != 200:
|
|
raise RuntimeError(f"Spark {r.status_code}: {r.text[:300]}")
|
|
data = r.json()
|
|
# 错误响应:{"error": {...}}
|
|
if "error" in data:
|
|
raise RuntimeError(f"Spark error: {data['error']}")
|
|
content = (
|
|
data.get("choices", [{}])[0]
|
|
.get("message", {})
|
|
.get("content", "")
|
|
.strip()
|
|
)
|
|
if not content:
|
|
raise RuntimeError(f"Spark empty content: {r.text[:300]}")
|
|
await asyncio.sleep(self.interval_sec)
|
|
return TranslationResult(
|
|
text=content, engine=self.name, chars=len(text), cached=False
|
|
)
|
|
except Exception as e:
|
|
last_exc = e
|
|
logger.warning("spark attempt %s failed: %s", attempt, e)
|
|
if attempt == 0:
|
|
await asyncio.sleep(0.5 + random.random())
|
|
else:
|
|
raise
|
|
assert last_exc is not None
|
|
raise last_exc
|