Merge branch 'main' into qbits_awq

Author: wenhuach21

auto_round/auto_quantizer.py

@@ -405,8 +405,8 @@ def convert_model(self, model: nn.Module):
                 logger.info(f"Change the dtype to `bfloat16` as {target_device.upper()} does not support float16")
                 model = model.to(torch.bfloat16)
             elif "cuda" == target_device and model.dtype != torch.float16:
-                    logger.info(f"Change the dtype to `float16` for better performance")
-                    model = model.to(torch.float16)
+                logger.info(f"Change the dtype to `float16` for better performance")
+                model = model.to(torch.float16)
 
         bits = quantization_config.bits
         group_size = quantization_config.group_size