Merge branch 'lora_modular' of https://gitee.com/lxdcumt/FlagScale_dev into lora_modular

Author: lxd-cumt

flagscale/train/peft/lora.py

@@ -30,7 +30,6 @@ class to provide a specific implementation of the forward method.
     def forward(
         self, x: torch.Tensor, *args, **kwargs
     ) -> tuple[torch.Tensor, Optional[torch.Tensor]]:
-        # pylint: disable=C0115,C0116
         linear_output, bias, layernorm_output = self.base_linear_forward(x, *args, **kwargs)
         adapter_output = self.adapter(layernorm_output.contiguous())
         adapter_output = adapter_output.reshape(linear_output.shape)