diff --git a/vllm/model_executor/layers/quantization/modelopt.py b/vllm/model_executor/layers/quantization/modelopt.py
index de0b0655ef..bac19536f4 100644
--- a/vllm/model_executor/layers/quantization/modelopt.py
+++ b/vllm/model_executor/layers/quantization/modelopt.py
@@ -450,7 +450,7 @@ class ModelOptFp8LinearMethod(LinearMethodBase):
 
     def __init__(self, quant_config: ModelOptFp8Config) -> None:
         self.quant_config = quant_config
-        self.out_dtype = torch.get_default_dtype()
+        self.out_dtype = get_current_vllm_config().model_config.dtype
         self.input_dtype = get_current_vllm_config().model_config.dtype
 
     def create_weights(
@@ -543,7 +543,7 @@ class ModelOptFp8PcPtLinearMethod(LinearMethodBase):
 
     def __init__(self, quant_config: ModelOptFp8Config) -> None:
         self.quant_config = quant_config
-        self.out_dtype = torch.get_default_dtype()
+        self.out_dtype = get_current_vllm_config().model_config.dtype
         self.input_dtype = get_current_vllm_config().model_config.dtype
 
     def create_weights(
@@ -640,7 +640,7 @@ class ModelOptFp8PbWoLinearMethod(LinearMethodBase):
             static=True, group_shape=GroupShape(block_n, block_k)
         )
 
-        self.out_dtype = torch.get_default_dtype()
+        self.out_dtype = get_current_vllm_config().model_config.dtype
         self.input_dtype = get_current_vllm_config().model_config.dtype
 
     def create_weights(