update

app.py

@@ -22,17 +22,21 @@ app = FastAPI()
 
 # 加载重排序模型（初始化时加载，避免每次请求重复加载）
 try:
-    model = CrossEncoder("BAAI/bge-reranker-large")  # 关键修改：使用 CrossEncoder 而非 SentenceTransformer
-    logger.info("Reranker model loaded successfully")
+    model = CrossEncoder(
+        "BAAI/bge-reranker-large",
+        tokenizer_args={"truncation": True},  # 在初始化时设置截断
+        max_length=512  # 可选：限制最大长度
+    )
+    logger.info("Model loaded with truncation support")
 except Exception as e:
-    logger.critical(f"Model loading failed: {e}")
-    raise RuntimeError("Failed to initialize model")
+    logger.critical(f"Model load failed: {str(e)}")
+    raise RuntimeError("Model initialization error")
 
 # 请求体模型
 class RerankRequest(BaseModel):
-    query: str = Field(..., min_length=1, max_length=8192, description="搜索query文本")
-    documents: List[str] = Field(..., min_items=1, description="待排序的文档列表")
-    top_k: int = Field(None, ge=1, le=100, description="返回前K个结果（默认返回全部）")
+    query: str = Field(..., min_length=1, max_length=8192)
+    documents: List[str] = Field(..., min_items=1)
+    top_k: int = Field(None, ge=1, le=100)
 
 # 响应模型
 class RerankResult(BaseModel):