diff --git a/diopi_test/python/configs/diopi_configs.py b/diopi_test/python/configs/diopi_configs.py
index 0ad81385e..39e9c4204 100755
--- a/diopi_test/python/configs/diopi_configs.py
+++ b/diopi_test/python/configs/diopi_configs.py
@@ -5552,8 +5552,8 @@
     'fused_adamw': dict(
          name=['fused_adamw'],
          interface=["CustomizedTest"],
-         atol=1e-2,
-         rtol=2e-3,
+         atol=3e-5,
+         rtol=3e-5,
          atol_half=1e-2,
          rtol_half=2e-3,
          para=dict(
diff --git a/diopi_test/python/conformance/customized_test.py b/diopi_test/python/conformance/customized_test.py
index 3f351e27c..62a4f9141 100644
--- a/diopi_test/python/conformance/customized_test.py
+++ b/diopi_test/python/conformance/customized_test.py
@@ -163,7 +163,7 @@ def fused_adamw(
         amsgrad,
         maximize,
     ):
-        torch.optim._functional.adamw(
+        torch._fused_adamw_(
             params,
             grads,
             exp_avgs,
@@ -171,13 +171,12 @@ def fused_adamw(
             max_exp_avg_sqs,
             state_steps,
             amsgrad=amsgrad,
+            lr=lr,
             beta1=beta1,
             beta2=beta2,
-            lr=lr,
             weight_decay=weight_decay,
             eps=eps,
             maximize=maximize,
-            fused=True,
         )
         return params, exp_avgs, exp_avg_sqs, max_exp_avg_sqs
 
diff --git a/impl/torch/functions/functions.cpp b/impl/torch/functions/functions.cpp
index 6ee8e104e..7e5a46111 100644
--- a/impl/torch/functions/functions.cpp
+++ b/impl/torch/functions/functions.cpp
@@ -3360,7 +3360,7 @@ diopiError_t diopiGridSample(diopiContextHandle_t ctx, diopiTensorHandle_t out,
 
 diopiError_t diopiFusedAdamW(diopiContextHandle_t ctx, diopiTensorHandle_t* params, diopiConstTensorHandle_t* grads, diopiTensorHandle_t* exp_avgs,
                              diopiTensorHandle_t* exp_avg_sqs, diopiTensorHandle_t* max_exp_avg_sqs, diopiConstTensorHandle_t* state_steps, int64_t nums,
-                             float lr, float beta1, float beta2, float eps, float weight_decay, bool amsgrad, bool maximize) {
+                             double lr, double beta1, double beta2, double eps, double weight_decay, bool amsgrad, bool maximize) {
     impl::aten::setCurStream(ctx);
     DIOPI_CHECK_PTR(params);
     DIOPI_IMPL_BUILD_ATEN_LIST(atParam, params, nums);
diff --git a/proto/include/diopi/functions.h b/proto/include/diopi/functions.h
index 4f7dfcecb..e66d12f5e 100644
--- a/proto/include/diopi/functions.h
+++ b/proto/include/diopi/functions.h
@@ -2829,7 +2829,7 @@ DIOPI_API diopiError_t diopiReciprocalInp(diopiContextHandle_t ctx, diopiTensorH
  */
 DIOPI_API diopiError_t diopiFusedAdamW(diopiContextHandle_t ctx, diopiTensorHandle_t* params, diopiConstTensorHandle_t* grads, diopiTensorHandle_t* exp_avgs,
                                        diopiTensorHandle_t* exp_avg_sqs, diopiTensorHandle_t* max_exp_avg_sqs, diopiConstTensorHandle_t* state_steps,
-                                       int64_t nums, float lr, float beta1, float beta2, float eps, float weight_decay, bool amsgrad, bool maximize);
+                                       int64_t nums, double lr, double beta1, double beta2, double eps, double weight_decay, bool amsgrad, bool maximize);
 
 /**
  * @brief The function is used to implement the AdamW optimizer. Its functionality is to perform a single parameter update.