diff --git a/mindspore/numpy/dlpack_1.py b/mindspore/numpy/dlpack_1.py
new file mode 100644
index 00000000000..f77daaa70bf
--- /dev/null
+++ b/mindspore/numpy/dlpack_1.py
@@ -0,0 +1,43 @@
+DLManagedTensor* toDLPack(constTensor& src){
+ATenDLMTensor * atDLMTensor(newATenDLMTensor);
+
+atDLMTensor->handle = src;
+
+atDLMTensor->tensor.manager_ctx = atDLMTensor;
+
+atDLMTensor->tensor.deleter = &deleter;
+
+atDLMTensor->tensor.dl_tensor.data = src.data_ptr();
+
+int64_tdevice_id = 0;
+
+if(src.type().is_cuda()) {
+device_id = src.get_device();
+
+}
+
+atDLMTensor->tensor.dl_tensor.ctx = getDLContext(src.type(), device_id);
+
+atDLMTensor->tensor.dl_tensor.ndim = src.dim(
+);
+atDLMTensor->tensor.dl_tensor.dtype = getDLDataType(src.type());
+
+atDLMTensor->tensor.dl_tensor.shape = const_cast< int64_t*>(src.sizes().data());
+
+atDLMTensor->tensor.dl_tensor.strides = const_cast< int64_t*>(src.strides().data());
+
+atDLMTensor->tensor.dl_tensor.byte_offset = 0;
+
+return&(atDLMTensor->tensor);
+
+}
+
+
+DLPack FromDLPack(DLtensor){
+DLtensor dltensor;
+dltensor.data* = mindsporeTensor.data;
+dltensor.ndim = mindsporeTensor.ndim;
+dltensor.dtype = mindsporeTensor.dtype;
+dltensor.shape* = mindsporeTensor.shape;
+dltensor.stride* = mindsporeTensor.stride;
+}