未验证 提交 0a0fbb1a 编写于 作者: C Chen Weihang 提交者: GitHub

polish tensor comment, test=document_fix (#54298)

上级 9189a6b4
...@@ -58,17 +58,13 @@ class AbstractAutogradMeta { ...@@ -58,17 +58,13 @@ class AbstractAutogradMeta {
/** /**
* Tensor is the API description of the basic data structure in the * Tensor is the API description of the basic data structure in the
* [ "Paddle Tensor Operation (phi)" Library ]. * [ "Paddle HIgh reusability operator (phi)" Library ].
* *
* It is not limited to a simple n-dimensional array. * It is not limited to a simple n-dimensional array.
* It contains a smart pointer to `TensorImpl`. The data description contained * It contains a smart pointer to `TensorImpl`. The data description contained
* in Tensor is defined by TensorImpl. Tensor only defines the interface for * in Tensor is defined by TensorImpl. Tensor only defines the interface for
* computation. * computation.
* *
* This is a new Tensor design, which is independent of the original
* phi::DenseTensor in fluid. The original Tensor will be gradually discarded
* in the future.
*
* Note: Tensor can be NULL state, Tensor is meaningful only when the * Note: Tensor can be NULL state, Tensor is meaningful only when the
* TensorImpl to which it is pointed is not empty. * TensorImpl to which it is pointed is not empty.
* *
...@@ -79,8 +75,8 @@ class AbstractAutogradMeta { ...@@ -79,8 +75,8 @@ class AbstractAutogradMeta {
* Note: Tensor cannot be inherited. The heterogeneous Tensor implementation * Note: Tensor cannot be inherited. The heterogeneous Tensor implementation
* can be achieved by inheriting the underlying TensorBase. * can be achieved by inheriting the underlying TensorBase.
* *
* Note: This Tensor API is suitable for training and custom operators, * Note: This Tensor API is suitable for training,inference and custom
* another simple Tensor design may be required for inference. * operators.
*/ */
class PADDLE_API Tensor final { class PADDLE_API Tensor final {
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册