The model is the output of training process. One complete model consists of two parts, namely, the **topology** and the **parameters**. To support industrial deployment, we need to make the model format must be self-completed and do not expose any training source code.
As a result, In PaddlePaddle, the **topology** represents as a [ProgramDesc](https://github.com/PaddlePaddle/Paddle/blob/1c0a4c901c9fc881d120249c703b15d1c50dae7d/doc/design/program.md), which describes the model structure. The **parameters** contain all the trainable weights in the model, we must support large size parameter, and efficient serialization/deserialization.
## Implementation
The topology is saved as a plain text, in detail, a self-contain protobuf file.
The parameters are saved as a binary file. As we all know, the protobuf message has the limits of [64M size](https://developers.google.com/protocol-buffers/docs/reference/cpp/google.protobuf.io.coded_stream#CodedInputStream.SetTotalBytesLimit.details). We do a (benchmark experiment)[https://github.com/PaddlePaddle/Paddle/pull/4610], its result shows protobuf is not fit in this scene.
As a result, we design a particular format for tensor serialization. By default, arbitrary tensor in Paddle is a [LoDTensor](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/lod_tensor.md), and has a description information proto of (LoDTensorDesc)[https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/framework.proto#L99]. We save the DescProto as the byte string header, it contains the necessary information, such as the `dims`, the `name` of the tensor, and the `LoD` information in [LoDTensor](https://github.com/PaddlePaddle/Paddle/blob/1c0a4c901c9fc881d120249c703b15d1c50dae7d/paddle/framework/lod_tensor.md). Tensor stores value in a continuous memory buffer, for speed we dump the raw memory to disk and save it as the byte string content. So, the binary format of one tensor is,
In detail, tensor's byte view as the table shows. Note that all the signed value written in little-endian.
```text
[offset] [type] [description]
0004 4 bytes integer HeaderLength, the length of LoDTensorDesc
0008 4 bytes integer ContentLength, the length of LodTensor Buffer
0009 1 bytes char TensorDesc
00010 1 bytes char TensorDesc
...
00100 1 bytes char TensorValue
00101 1 bytes char TensorValue
00102 1 bytes char TensorValue ..
...
```
## Summary
We introduce the model format, the `ProgramDesc` describe the **topology**, and a bunch of particular format binary tensors describes the **parameters**.
<liclass="toctree-l2"><aclass="reference internal"href="../getstarted/build_and_install/index_en.html">Install and Build</a><ul>
<liclass="toctree-l3"><aclass="reference internal"href="../getstarted/build_and_install/docker_install_en.html">PaddlePaddle in Docker Containers</a></li>
<liclass="toctree-l3"><aclass="reference internal"href="../getstarted/build_and_install/build_from_source_en.html">Installing from Sources</a></li>
<liclass="toctree-l2"><aclass="reference internal"href="../howto/usage/cluster/cluster_train_en.html#use-cluster-platforms-or-cluster-management-tools">Use cluster platforms or cluster management tools</a></li>
<liclass="toctree-l2"><aclass="reference internal"href="../howto/usage/k8s/k8s_en.html">Paddle On Kubernetes</a></li>
<liclass="toctree-l2"><aclass="reference internal"href="../howto/usage/k8s/k8s_aws_en.html">Distributed PaddlePaddle Training on AWS with Kubernetes</a></li>
<liclass="toctree-l2"><aclass="reference internal"href="../howto/dev/build_en.html">Build PaddlePaddle from Source Code and Run Unit Test</a></li>
<liclass="toctree-l2"><aclass="reference internal"href="../howto/dev/new_layer_en.html">Write New Layers</a></li>
<spanid="design-doc-model-format"></span><h1>Design Doc: Model Format<aclass="headerlink"href="#design-doc-model-format"title="Permalink to this headline">¶</a></h1>
<divclass="section"id="motivation">
<spanid="motivation"></span><h2>Motivation<aclass="headerlink"href="#motivation"title="Permalink to this headline">¶</a></h2>
<p>The model is the output of training process. One complete model consists of two parts, namely, the <strong>topology</strong> and the <strong>parameters</strong>. To support industrial deployment, we need to make the model format must be self-completed and do not expose any training source code.</p>
<p>As a result, In PaddlePaddle, the <strong>topology</strong> represents as a <aclass="reference external"href="https://github.com/PaddlePaddle/Paddle/blob/1c0a4c901c9fc881d120249c703b15d1c50dae7d/doc/design/program.md">ProgramDesc</a>, which describes the model structure. The <strong>parameters</strong> contain all the trainable weights in the model, we must support large size parameter, and efficient serialization/deserialization.</p>
</div>
<divclass="section"id="implementation">
<spanid="implementation"></span><h2>Implementation<aclass="headerlink"href="#implementation"title="Permalink to this headline">¶</a></h2>
<p>The topology is saved as a plain text, in detail, a self-contain protobuf file.</p>
<p>The parameters are saved as a binary file. As we all know, the protobuf message has the limits of <aclass="reference external"href="https://developers.google.com/protocol-buffers/docs/reference/cpp/google.protobuf.io.coded_stream#CodedInputStream.SetTotalBytesLimit.details">64M size</a>. We do a (benchmark experiment)[https://github.com/PaddlePaddle/Paddle/pull/4610], its result shows protobuf is not fit in this scene.</p>
<p>As a result, we design a particular format for tensor serialization. By default, arbitrary tensor in Paddle is a <aclass="reference external"href="https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/lod_tensor.md">LoDTensor</a>, and has a description information proto of (LoDTensorDesc)[https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/framework.proto#L99]. We save the DescProto as the byte string header, it contains the necessary information, such as the <codeclass="docutils literal"><spanclass="pre">dims</span></code>, the <codeclass="docutils literal"><spanclass="pre">name</span></code> of the tensor, and the <codeclass="docutils literal"><spanclass="pre">LoD</span></code> information in <aclass="reference external"href="https://github.com/PaddlePaddle/Paddle/blob/1c0a4c901c9fc881d120249c703b15d1c50dae7d/paddle/framework/lod_tensor.md">LoDTensor</a>. Tensor stores value in a continuous memory buffer, for speed we dump the raw memory to disk and save it as the byte string content. So, the binary format of one tensor is,</p>
0004 4 bytes integer HeaderLength, the length of LoDTensorDesc
0008 4 bytes integer ContentLength, the length of LodTensor Buffer
0009 1 bytes char TensorDesc
00010 1 bytes char TensorDesc
...
00100 1 bytes char TensorValue
00101 1 bytes char TensorValue
00102 1 bytes char TensorValue ..
...
</pre></div>
</div>
</div>
<divclass="section"id="summary">
<spanid="summary"></span><h2>Summary<aclass="headerlink"href="#summary"title="Permalink to this headline">¶</a></h2>
<p>We introduce the model format, the <codeclass="docutils literal"><spanclass="pre">ProgramDesc</span></code> describe the <strong>topology</strong>, and a bunch of particular format binary tensors describes the <strong>parameters</strong>.</p>
Built with <ahref="http://sphinx-doc.org/">Sphinx</a> using a <ahref="https://github.com/snide/sphinx_rtd_theme">theme</a> provided by <ahref="https://readthedocs.org">Read the Docs</a>.
The model is the output of training process. One complete model consists of two parts, namely, the **topology** and the **parameters**. To support industrial deployment, we need to make the model format must be self-completed and do not expose any training source code.
As a result, In PaddlePaddle, the **topology** represents as a [ProgramDesc](https://github.com/PaddlePaddle/Paddle/blob/1c0a4c901c9fc881d120249c703b15d1c50dae7d/doc/design/program.md), which describes the model structure. The **parameters** contain all the trainable weights in the model, we must support large size parameter, and efficient serialization/deserialization.
## Implementation
The topology is saved as a plain text, in detail, a self-contain protobuf file.
The parameters are saved as a binary file. As we all know, the protobuf message has the limits of [64M size](https://developers.google.com/protocol-buffers/docs/reference/cpp/google.protobuf.io.coded_stream#CodedInputStream.SetTotalBytesLimit.details). We do a (benchmark experiment)[https://github.com/PaddlePaddle/Paddle/pull/4610], its result shows protobuf is not fit in this scene.
As a result, we design a particular format for tensor serialization. By default, arbitrary tensor in Paddle is a [LoDTensor](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/lod_tensor.md), and has a description information proto of (LoDTensorDesc)[https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/framework.proto#L99]. We save the DescProto as the byte string header, it contains the necessary information, such as the `dims`, the `name` of the tensor, and the `LoD` information in [LoDTensor](https://github.com/PaddlePaddle/Paddle/blob/1c0a4c901c9fc881d120249c703b15d1c50dae7d/paddle/framework/lod_tensor.md). Tensor stores value in a continuous memory buffer, for speed we dump the raw memory to disk and save it as the byte string content. So, the binary format of one tensor is,
In detail, tensor's byte view as the table shows. Note that all the signed value written in little-endian.
```text
[offset] [type] [description]
0004 4 bytes integer HeaderLength, the length of LoDTensorDesc
0008 4 bytes integer ContentLength, the length of LodTensor Buffer
0009 1 bytes char TensorDesc
00010 1 bytes char TensorDesc
...
00100 1 bytes char TensorValue
00101 1 bytes char TensorValue
00102 1 bytes char TensorValue ..
...
```
## Summary
We introduce the model format, the `ProgramDesc` describe the **topology**, and a bunch of particular format binary tensors describes the **parameters**.
<p>The model is the output of training process. One complete model consists of two parts, namely, the <strong>topology</strong> and the <strong>parameters</strong>. To support industrial deployment, we need to make the model format must be self-completed and do not expose any training source code.</p>
<p>As a result, In PaddlePaddle, the <strong>topology</strong> represents as a <aclass="reference external"href="https://github.com/PaddlePaddle/Paddle/blob/1c0a4c901c9fc881d120249c703b15d1c50dae7d/doc/design/program.md">ProgramDesc</a>, which describes the model structure. The <strong>parameters</strong> contain all the trainable weights in the model, we must support large size parameter, and efficient serialization/deserialization.</p>
<p>The topology is saved as a plain text, in detail, a self-contain protobuf file.</p>
<p>The parameters are saved as a binary file. As we all know, the protobuf message has the limits of <aclass="reference external"href="https://developers.google.com/protocol-buffers/docs/reference/cpp/google.protobuf.io.coded_stream#CodedInputStream.SetTotalBytesLimit.details">64M size</a>. We do a (benchmark experiment)[https://github.com/PaddlePaddle/Paddle/pull/4610], its result shows protobuf is not fit in this scene.</p>
<p>As a result, we design a particular format for tensor serialization. By default, arbitrary tensor in Paddle is a <aclass="reference external"href="https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/lod_tensor.md">LoDTensor</a>, and has a description information proto of (LoDTensorDesc)[https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/framework.proto#L99]. We save the DescProto as the byte string header, it contains the necessary information, such as the <codeclass="docutils literal"><spanclass="pre">dims</span></code>, the <codeclass="docutils literal"><spanclass="pre">name</span></code> of the tensor, and the <codeclass="docutils literal"><spanclass="pre">LoD</span></code> information in <aclass="reference external"href="https://github.com/PaddlePaddle/Paddle/blob/1c0a4c901c9fc881d120249c703b15d1c50dae7d/paddle/framework/lod_tensor.md">LoDTensor</a>. Tensor stores value in a continuous memory buffer, for speed we dump the raw memory to disk and save it as the byte string content. So, the binary format of one tensor is,</p>
<p>We introduce the model format, the <codeclass="docutils literal"><spanclass="pre">ProgramDesc</span></code> describe the <strong>topology</strong>, and a bunch of particular format binary tensors describes the <strong>parameters</strong>.</p>
Built with <ahref="http://sphinx-doc.org/">Sphinx</a> using a <ahref="https://github.com/snide/sphinx_rtd_theme">theme</a> provided by <ahref="https://readthedocs.org">Read the Docs</a>.
