quick fix for dsv3 perchannel/pertensor (online quantization) (#774)

Co-authored-by: shihaobai <[email protected]>

Author: shihaobai

lightllm/common/quantization/quantize_method.py

@@ -7,6 +7,8 @@ class QuantizationMethod(ABC):
     def __init__(self):
         super().__init__()
         self.device_id_ = get_current_device_id()
+        self.weight_scale_suffix = None
+        self.act_scale_suffix = None
 
     @abstractmethod
     def quantize(self, weights: torch.Tensor):