Update localrerank.py
Browse files- localrerank.py +3 -3
localrerank.py
CHANGED
|
@@ -13,7 +13,7 @@ security = HTTPBearer()
|
|
| 13 |
|
| 14 |
SK_KEY = os.environ.get("SK_KEY", "sk-aaabbbcccdddeeefffggghhhiiijjjkkk")
|
| 15 |
MODEL_ID = os.environ.get("RERANK_MODEL", "Qwen/Qwen3-Reranker-4B")
|
| 16 |
-
MAX_LENGTH = int(os.environ.get("MAX_LENGTH", "
|
| 17 |
DEVICE = 'cuda' if torch.cuda.is_available() else 'cpu'
|
| 18 |
|
| 19 |
model = None
|
|
@@ -26,8 +26,8 @@ class RerankRequest(BaseModel):
|
|
| 26 |
)
|
| 27 |
query: str
|
| 28 |
documents: Union[List[str], str]
|
| 29 |
-
top_k: int = Field(default=5, ge=1, le=
|
| 30 |
-
batch_size: int = Field(default=
|
| 31 |
return_documents: bool = True
|
| 32 |
|
| 33 |
@validator("documents", pre=True)
|
|
|
|
| 13 |
|
| 14 |
SK_KEY = os.environ.get("SK_KEY", "sk-aaabbbcccdddeeefffggghhhiiijjjkkk")
|
| 15 |
MODEL_ID = os.environ.get("RERANK_MODEL", "Qwen/Qwen3-Reranker-4B")
|
| 16 |
+
MAX_LENGTH = int(os.environ.get("MAX_LENGTH", "512"))
|
| 17 |
DEVICE = 'cuda' if torch.cuda.is_available() else 'cpu'
|
| 18 |
|
| 19 |
model = None
|
|
|
|
| 26 |
)
|
| 27 |
query: str
|
| 28 |
documents: Union[List[str], str]
|
| 29 |
+
top_k: int = Field(default=5, ge=1, le=50)
|
| 30 |
+
batch_size: int = Field(default=4, ge=1, le=32)
|
| 31 |
return_documents: bool = True
|
| 32 |
|
| 33 |
@validator("documents", pre=True)
|