diff --git a/src/backend/cuda/operator/LSQImpl.cpp b/src/backend/cuda/operator/LSQImpl.cpp
index c66bd8a5aa78513b4bcceec83f9c9d87ffed2b11..fa45f211e72f6742b72584aadf2a109c3bdca594 100644
--- a/src/backend/cuda/operator/LSQImpl.cpp
+++ b/src/backend/cuda/operator/LSQImpl.cpp
@@ -52,19 +52,6 @@ void Aidge::LSQImpl_cuda::backward() {
     std::shared_ptr<Tensor> gra_int1 = op_.getInput(1)->grad();
     std::shared_ptr<Tensor> gra_out0 = op_.getOutput(0)->grad();    
 
-    // XXX
-/*
-    size_t tmp;
-
-    cudaDeviceSetLimit(cudaLimitStackSize, 2048);
-    cudaDeviceGetLimit(&tmp, cudaLimitStackSize );
-    printf(" stack limit = %ld \n", tmp);
-
-    cudaDeviceSetLimit(cudaLimitMallocHeapSize, 100000000);
-    cudaDeviceGetLimit(&tmp, cudaLimitMallocHeapSize);
-    printf(" heap limit = %ld \n", tmp);
-*/
-
     if (gra_int0->size() > mWorkspaceSize) {
         // std::cout << " reallocation " << sizeof(gra_int0) << " " << gra_int0->size() << std::endl;
         if (mWorkspace != nullptr) {
@@ -87,12 +74,7 @@ void Aidge::LSQImpl_cuda::backward() {
         gra_int0->getImpl()->rawPtr(),
         gra_int1->getImpl()->rawPtr(),
         mWorkspace);
-/*
-    gra_int1->setBackend("cpu");
-    float *castedTensor = static_cast<float *> (gra_int1->getImpl()->rawPtr());
-    std::cout << castedTensor[0] << std::endl;
-    gra_int1->setBackend("cuda");
-*/
+
 }
 
 Aidge::LSQImpl_cuda::~LSQImpl_cuda() {