diff --git a/python/jittor/__init__.py b/python/jittor/__init__.py
index f224021f..23124737 100644
--- a/python/jittor/__init__.py
+++ b/python/jittor/__init__.py
@@ -9,7 +9,7 @@
 # file 'LICENSE.txt', which is part of this source code package.
 # ***************************************************************
 
-__version__ = '1.3.2.5'
+__version__ = '1.3.2.6'
 from jittor_utils import lock
 with lock.lock_scope():
     ori_int = int
diff --git a/python/jittor/src/executor.cc b/python/jittor/src/executor.cc
index 186eeb0c..5deacd53 100644
--- a/python/jittor/src/executor.cc
+++ b/python/jittor/src/executor.cc
@@ -499,7 +499,8 @@ void Executor::run_sync(vector<Var*> vars, bool device_sync) {
                 sync_times++;
             }
             for (Var* v : op->inputs()) {
-                migrate_to_cpu(v, allocator);
+                if (v->allocator->is_cuda())
+                    migrate_to_cpu(v, allocator);
             }
             if (!use_cuda_managed_allocator) {
                 for (auto* var : op->outputs()) {
diff --git a/python/jittor/src/mem/allocator.cc b/python/jittor/src/mem/allocator.cc
index 1e425b4f..46c6f308 100644
--- a/python/jittor/src/mem/allocator.cc
+++ b/python/jittor/src/mem/allocator.cc
@@ -105,6 +105,7 @@ void migrate_to_cpu(Var* var, Allocator* allocator) {
         );
     } else
     if (!use_cuda_managed_allocator) {
+        if (!var->allocator->is_cuda()) return;
         // must be a device allocator
         Allocation a(allocator, var->size);
         checkCudaErrors(cudaMemcpy(a.ptr, var->mem_ptr, var->size, cudaMemcpyDeviceToHost));