diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml
index ff2c6e2b..13d94eb2 100644
--- a/.github/workflows/release.yml
+++ b/.github/workflows/release.yml
@@ -5,6 +5,7 @@ on:
   push:
     branches:
       - main
+      - debug
     tags:
       - v*
 
@@ -34,20 +35,3 @@ jobs:
         with:
           name: wheel
           path: wheelhouse/
-
-  publish:
-    runs-on: ubuntu-latest
-    needs: [release]
-    steps:
-      - uses: actions/download-artifact@v3
-        with:
-          path: artifact
-      - name: Move files so the next action can find them
-        run: |
-          mkdir dist && mv artifact/wheel/* dist/
-          ls dist/
-      - name: Publish distribution to PyPI
-        if: startsWith(github.ref, 'refs/tags')
-        uses: pypa/gh-action-pypi-publish@release/v1
-        with:
-          password: ${{ secrets.PYPI_API_TOKEN }}
diff --git a/envpool/core/xla.h b/envpool/core/xla.h
index e70fbb98..f4d9afff 100644
--- a/envpool/core/xla.h
+++ b/envpool/core/xla.h
@@ -74,8 +74,8 @@ Array GpuBufferToArray(cudaStream_t stream, const void* buffer,
     spec = spec.Batch(batch_size);
   }
   Array ret(spec);
-  cudaMemcpy(ret.Data(), buffer, ret.size * ret.element_size,
-             cudaMemcpyDeviceToHost);
+  cudaMemcpyAsync(ret.Data(), buffer, ret.size * ret.element_size,
+                  cudaMemcpyDeviceToHost, stream);
   return ret;
 }
 
@@ -161,6 +161,7 @@ struct XlaSend {
            ...);
         },
         action_spec);
+    cudaStreamSynchronize(stream);
     envpool->Send(action);
   }
 };
diff --git a/envpool/core/xla_template.h b/envpool/core/xla_template.h
index 2da813db..36e4226c 100644
--- a/envpool/core/xla_template.h
+++ b/envpool/core/xla_template.h
@@ -26,6 +26,7 @@
 #include <string>
 #include <tuple>
 #include <vector>
+#include <cassert>
 
 namespace py = pybind11;
 
@@ -36,6 +37,9 @@ static auto SpecToTuple(const Spec& spec) {
 
 template <std::size_t N>
 void ToArray(const void** raw, std::array<void*, N>* array) {
+  for (int j = 0; j < N; ++j) {
+    assert(raw[j] != nullptr);
+  }
   int i = 0;
   std::apply([&](auto&&... a) { ((a = const_cast<void*>(raw[i++])), ...); },
              *array);
@@ -43,6 +47,9 @@ void ToArray(const void** raw, std::array<void*, N>* array) {
 
 template <std::size_t N>
 void ToArray(void** raw, std::array<void*, N>* array) {
+  for (int j = 0; j < N; ++j) {
+    assert(raw[j] != nullptr);
+  }
   int i = 0;
   std::apply([&](auto&&... a) { ((a = raw[i++]), ...); }, *array);
 }