@@ -7,7 +7,7 @@ Polyak and Juditsky (1992) showed that the test performance of simple average of
...
@@ -7,7 +7,7 @@ Polyak and Juditsky (1992) showed that the test performance of simple average of
Hence, to accelerate the speed of Stochastic Gradient Descent, Averaged Stochastic Gradient Descent (ASGD) was proposed in Polyak and Juditsky (1992). For ASGD, the running average of parameters obtained by SGD, is used as the estimator for <imgsrc="./images/theta_star.gif"/><br/> . The averaging is done as follows:
Hence, to accelerate the speed of Stochastic Gradient Descent, Averaged Stochastic Gradient Descent (ASGD) was proposed in Polyak and Juditsky (1992). For ASGD, the running average of parameters obtained by SGD, is used as the estimator for <imgsrc="./images/theta_star.gif"/><br/> . The averaging is done as follows:
<imgsrc="./images/asgd.gif"align="center"/><br/>
We propose averaging for any optimizer similar to how ASGD performs it, as mentioned above.
We propose averaging for any optimizer similar to how ASGD performs it, as mentioned above.
@@ -14,11 +14,29 @@ In programming languages, a block is a pair of curly braces that includes local
...
@@ -14,11 +14,29 @@ In programming languages, a block is a pair of curly braces that includes local
Blocks work with control flow structures like `if`, `else`, and `for`, which have equivalents in deep learning:
Blocks work with control flow structures like `if`, `else`, and `for`, which have equivalents in deep learning:
| programming languages | PaddlePaddle |
<table>
|-----------------------|-----------------------|
<thead>
| for, while loop | RNN, WhileOp |
<tr>
| if, if-else, switch | IfElseOp, SwitchOp |
<th>programming languages</th>
| sequential execution | a sequence of layers |
<th>PaddlePaddle</th>
</tr>
</thead>
<tbody>
<tr>
<td>for, while loop </td>
<td>RNN, WhileOp </td>
</tr>
<tr>
<td>if, if-else, switch </td>
<td>IfElseOp, SwitchOp </td>
</tr>
<tr>
<td>sequential execution </td>
<td>a sequence of layers </td>
</tr>
</tbody>
</table>
A key difference is that a C++ program describes a one pass computation, whereas a deep learning program describes both the forward and backward passes.
A key difference is that a C++ program describes a one pass computation, whereas a deep learning program describes both the forward and backward passes.
...
@@ -26,12 +44,33 @@ A key difference is that a C++ program describes a one pass computation, whereas
...
@@ -26,12 +44,33 @@ A key difference is that a C++ program describes a one pass computation, whereas
The existence of the backward pass makes the execution of a block of PaddlePaddle different from traditional programs:
The existence of the backward pass makes the execution of a block of PaddlePaddle different from traditional programs:
We'd like to have Python bindings to operators in package `paddle.operator`, and Python compositions of operators in package `paddle.layer`. So we have the following concepts in above illustrative example:
We'd like to have Python bindings to operators in package `paddle.operator`, and Python compositions of operators in package `paddle.layer`. So we have the following concepts in above illustrative example:
Like other deep learning systems, PaddlePaddle supports training models from sequence data. Also, like other systems, PaddlePaddle represent a mini-batch of sequences as a Tensor. What is different is that PaddlePaddle doesn't require all sequences in a mini-batch to be of the same length. Thus no need for padding zeros.
Like other deep learning systems, PaddlePaddle supports training models from sequence data. Also, like other systems, PaddlePaddle represent a mini-batch of sequences as a Tensor. What is different is that PaddlePaddle doesn't require all sequences in a mini-batch to be of the same length. Thus no need for padding zeros.
PaddlePaddle achieves this flexibility by passing through a new data type, *LoD Tensor*, which is a Tensor attached with segmentation index known as *LoD*, between operators. The LoD index doesn't only segment a tensor, but also recursively segments sub-sequences. This document presents the design of LoD and LoDTensor.
PaddlePaddle achieves this flexibility by passing through a new data type, *LoD Tensor*, which is a Tensor attached with segmentation index known as *LoD*, between operators. The LoD index doesn't only segment a tensor, but also recursively segments sub-sequences. This document presents the design of LoD and LoDTensor.
@@ -10,10 +10,27 @@ PaddlePaddle uses proto message to describe compile time program because :
...
@@ -10,10 +10,27 @@ PaddlePaddle uses proto message to describe compile time program because :
The computation `Program` consists of nested `Blocks`. Each `Block` will consist of data(i.e. `Variable`) and `Operations`. The concept to represent them is in the table below.
The computation `Program` consists of nested `Blocks`. Each `Block` will consist of data(i.e. `Variable`) and `Operations`. The concept to represent them is in the table below.
@@ -116,7 +142,7 @@ An explanation of the above program:
...
@@ -116,7 +142,7 @@ An explanation of the above program:
-`fluid.k8s` is a package that provides access to Kubernetes API.
-`fluid.k8s` is a package that provides access to Kubernetes API.
-`fluid.k8s.get_worker_addrs` returns the list of IP and ports of all pods of the current job except for the current one (the master pod).
-`fluid.k8s.get_worker_addrs` returns the list of IP and ports of all pods of the current job except for the current one (the master pod).
-`fluid.tensor_array` creates a [tensor array](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/lod_tensor_array.h). `fluid.parallel_for` creates a `ParallelFor` intrinsic, which, when executed,
-`fluid.tensor_array` creates a [tensor array](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/lod_tensor_array.h). `fluid.parallel_for` creates a `ParallelFor` intrinsic, which, when executed,
1. creates `len(L)` scopes, each for the concurrent running of the sub-block (block 1 in this case), and initializes a variable named "index" in the scope to an integer value in the range `[0, len(L)-1]`, and
1. creates `len(L)` scopes, each for the concurrent running of the sub-block (block 1 in this case), and initializes a variable named "index" in the scope to an integer value in the range `[0, len(L)-1]`, and
2. creates `len(L)` threads by calling into the `ThreadPool` singleton, each thread
2. creates `len(L)` threads by calling into the `ThreadPool` singleton, each thread
@@ -10,11 +10,37 @@ Fluid is the answer. Fluid is similar to PyTorch and TensorFlow Eager Execution
...
@@ -10,11 +10,37 @@ Fluid is the answer. Fluid is similar to PyTorch and TensorFlow Eager Execution
Deep learning infrastructure is one of the fastest evolving technologies. Within four years, there have already been three generations of technologies invented.
Deep learning infrastructure is one of the fastest evolving technologies. Within four years, there have already been three generations of technologies invented.
| Existed since | model as sequence of layers | model as graph of operators | No model |
From the above table, we see that the deep learning technology is evolving towards getting rid of the concept of a model. To understand the reasons behind this direction, a comparison of the *programming paradigms* or the ways to program deep learning applications using these systems, would be helpful. The following section goes over these.
From the above table, we see that the deep learning technology is evolving towards getting rid of the concept of a model. To understand the reasons behind this direction, a comparison of the *programming paradigms* or the ways to program deep learning applications using these systems, would be helpful. The following section goes over these.
The word *graph* is interchangeable with *block* in this document. A graph consists of computation steps and local variables similar to a C++/Java program block, or a pair of parentheses(`{` and `}`).
The word *graph* is interchangeable with *block* in this document. A graph consists of computation steps and local variables similar to a C++/Java program block, or a pair of parentheses(`{` and `}`).
We are planning to build Deep Speech 2 (DS2) \[[1](#references)\], a powerful Automatic Speech Recognition (ASR) engine, on PaddlePaddle. For the first-stage plan, we have the following short-term goals:
We are planning to build Deep Speech 2 (DS2) \[[1](#references)\], a powerful Automatic Speech Recognition (ASR) engine, on PaddlePaddle. For the first-stage plan, we have the following short-term goals:
...
@@ -68,11 +68,33 @@ We roughly break down the project into 14 tasks:
...
@@ -68,11 +68,33 @@ We roughly break down the project into 14 tasks:
Phase I | Simplified model & components | *Task 1* ~ *Task 8*
<tr>
Phase II | Standard model & benchmarking & profiling | *Task 9* ~ *Task 12*
<th>Roadmap</th>
Phase III | Documentations | *Task13* ~ *Task14*
<th>Description</th>
<th> Parallelizable Tasks</th>
</tr>
</thead>
<tbody>
<tr>
<td>Phase I </td>
<td>Simplified model & components </td>
<td>Task 1 ~ Task 8</td>
</tr>
<tr>
<td>Phase II </td>
<td> Standard model & benchmarking & profiling</td>
<td>Task 9 ~ Task 12 </td>
</tr>
<tr>
<td>Phase III </td>
<td> Documentations</td>
<td> Task13 ~ Task14 </td>
</tr>
</tbody>
</table>
Issue for each task will be created later. Contributions, discussions and comments are all highly appreciated and welcomed!
Issue for each task will be created later. Contributions, discussions and comments are all highly appreciated and welcomed!
...
@@ -102,37 +124,82 @@ We don't have to persist on this 2-3-7-1-1-1 depth \[[2](#references)\]. Similar
...
@@ -102,37 +124,82 @@ We don't have to persist on this 2-3-7-1-1-1 depth \[[2](#references)\]. Similar
Key ingredients about the layers:
Key ingredients about the layers:
-**Data Layers**:
-**Data Layers**:
- Frame sequences data of audio **spectrogram** (with FFT).
- Frame sequences data of audio **spectrogram** (with FFT).
- Token sequences data of **transcription** text (labels).
- Token sequences data of **transcription** text (labels).
- These two type of sequences do not have the same lengthes, thus a CTC-loss layer is required.
- These two type of sequences do not have the same lengthes, thus a CTC-loss layer is required.
-**2D Convolution Layers**:
-**2D Convolution Layers**:
- Not only temporal convolution, but also **frequency convolution**. Like a 2D image convolution, but with a variable dimension (i.e. temporal dimension).
- Not only temporal convolution, but also **frequency convolution**. Like a 2D image convolution, but with a variable dimension (i.e. temporal dimension).
- With striding for only the first convlution layer.
- With striding for only the first convlution layer.
- No pooling for all convolution layers.
- No pooling for all convolution layers.
-**Uni-directional RNNs**
-**Uni-directional RNNs**
- Uni-directional + row convolution: for low-latency inference.
- Uni-directional + row convolution: for low-latency inference.
- Bi-direcitional + without row convolution: if we don't care about the inference latency.
- Bi-direcitional + without row convolution: if we don't care about the inference latency.
-**Row convolution**:
-**Row convolution**:
- For looking only a few steps ahead into the feature, instead of looking into a whole sequence in bi-directional RNNs.
- For looking only a few steps ahead into the feature, instead of looking into a whole sequence in bi-directional RNNs.
- Not nessesary if with bi-direcitional RNNs.
- Not nessesary if with bi-direcitional RNNs.
- "**Row**" means convolutions are done within each frequency dimension (row), and no convolution kernels shared across.
- "**Row**" means convolutions are done within each frequency dimension (row), and no convolution kernels shared across.
-**Batch Normalization Layers**:
-**Batch Normalization Layers**:
- Added to all above layers (except for data and loss layer).
- Added to all above layers (except for data and loss layer).
- Sequence-wise normalization for RNNs: BatchNorm only performed on input-state projection and not state-state projection, for efficiency consideration.
- Sequence-wise normalization for RNNs: BatchNorm only performed on input-state projection and not state-state projection, for efficiency consideration.
<table>
Required Components | PaddlePaddle Support | Need to Develop
CTC-Beam search | Not supported yet. | TBD (Task 6)
<td>Not supported yet.</td>
<td>TBD (Task 3)</td>
</tr>
<tr>
<td>Data Layer II (Transcription) </td>
<td> paddle.data_type.integer_value_sequence</td>
<td> - </td>
</tr>
<tr>
<td>2D Convolution Layer </td>
<td> paddle.layer.image_conv_layer</td>
<td> - </td>
</tr>
<tr>
<td>DataType Converter (vec2seq)</td>
<td> paddle.layer.block_expand</td>
<td> - </td>
</tr>
<tr>
<td>Bi-/Uni-directional RNNs </td>
<td>paddle.layer.recurrent_group</td>
<td> - </td>
</tr>
<tr>
<td>Row Convolution Layer </td>
<td>Not supported yet.</td>
<td>TBD (Task 4)</td>
</tr>
<tr>
<td>CTC-loss Layer </td>
<td>paddle.layer.warp_ctc</td>
<td> - </td>
</tr>
<tr>
<td>Batch Normalization Layer </td>
<td>paddle.layer.batch_norm</td>
<td> - </td>
</tr>
<tr>
<td>CTC-Beam search </td>
<td>Not supported yet.</td>
<td> TBD (Task 6) </td>
</tr>
</tbody>
</table>
### Row Convolution
### Row Convolution
...
@@ -145,14 +212,14 @@ TODO by Assignees
...
@@ -145,14 +212,14 @@ TODO by Assignees
Figure 2. Algorithm for CTC Beam Search Decoder.
Figure 2. Algorithm for CTC Beam Search Decoder.
</div>
</div>
- The **Beam Search Decoder** for DS2 CTC-trained network follows the similar approach in \[[3](#references)\] as shown in Figure 2, with two important modifications for the ambiguous parts:
- The **Beam Search Decoder** for DS2 CTC-trained network follows the similar approach in \[[3](#references)\] as shown in Figure 2, with two important modifications for the ambiguous parts:
- 1) in the iterative computation of probabilities, the assignment operation is changed to accumulation for one prefix may comes from different paths;
- 1) in the iterative computation of probabilities, the assignment operation is changed to accumulation for one prefix may comes from different paths;
- 2) the if condition ```if l^+ not in A_prev then``` after probabilities' computation is deprecated for it is hard to understand and seems unnecessary.
- 2) the if condition ```if l^+ not in A_prev then``` after probabilities' computation is deprecated for it is hard to understand and seems unnecessary.
- An **external scorer** would be passed into the decoder to evaluate a candidate prefix during decoding whenever a white space appended in English decoding and any character appended in Mandarin decoding.
- An **external scorer** would be passed into the decoder to evaluate a candidate prefix during decoding whenever a white space appended in English decoding and any character appended in Mandarin decoding.
- Such external scorer consists of language model, word count or any other custom scorers.
- Such external scorer consists of language model, word count or any other custom scorers.
- The **language model** is built from Task 5, with parameters should be carefully tuned to achieve minimum WER/CER (c.f. Task 7)
- The **language model** is built from Task 5, with parameters should be carefully tuned to achieve minimum WER/CER (c.f. Task 7)
- This decoder needs to perform with **high efficiency** for the convenience of parameters tuning and speech recognition in reality.
- This decoder needs to perform with **high efficiency** for the convenience of parameters tuning and speech recognition in reality.
Kernel implementation | The kernel methods shared between CPU and CUDA are defined in `.h` files. CPU-specific kernels live in `.cc` files, while CUDA-specific kernels are implemented in `.cu`files.
Kernel implementation | The kernel methods shared between CPU and CUDA are defined in `.h` files. CPU-specific kernels live in `.cc` files, while CUDA-specific kernels are implemented in `.cu`files.
Registering the Op | Ops are registered in `.cc` files; For Kernel registration, `.cc` files contain the CPU implementation, while `.cu` files contain the CUDA implementation.
Registering the Op | Ops are registered in `.cc` files; For Kernel registration, `.cc` files contain the CPU implementation, while `.cu` files contain the CUDA implementation.
<table>
<thead>
<tr>
<th>Information</th>
<th> Where is it defined</th>
</tr>
</thead>
<tbody>
<tr>
<td>OpProtoMake definition </td>
<td>`.cc`files, Backward Op does not need an OpProtoMake interface. </td>
</tr>
<tr>
<td>Op definition </td>
<td>`.cc` files</td>
</tr>
<tr>
<td>Kernel implementation </td>
<td> The kernel methods shared between CPU and CUDA are defined in `.h` files. CPU-specific kernels live in `.cc` files, while CUDA-specific kernels are implemented in `.cu`files.</td>
</tr>
<tr>
<td>Registering the Op </td>
<td> Ops are registered in `.cc` files; For Kernel registration, `.cc` files contain the CPU implementation, while `.cu` files contain the CUDA implementation.</td>
</tr>
</tbody>
</table>
New Operator implementations are added to the list [paddle/operators](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/operators), with file names in the format `*_op.h` (if applicable), `*_op.cc`, `*_op.cu` (if applicable).** The system will use the naming scheme to automatically build operators and their corresponding Python extensions.**
New Operator implementations are added to the list [paddle/operators](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/operators), with file names in the format `*_op.h` (if applicable), `*_op.cc`, `*_op.cu` (if applicable).** The system will use the naming scheme to automatically build operators and their corresponding Python extensions.**
...
@@ -279,7 +306,7 @@ A forward operator unit test inherits `unittest.TestCase` and defines metaclass
...
@@ -279,7 +306,7 @@ A forward operator unit test inherits `unittest.TestCase` and defines metaclass
A model is an output of the training process. One complete model consists of two parts, the **topology** and the **parameters**. In order to support industrial deployment, the model format must be self-complete and must not expose any training source code.
A model is an output of the training process. One complete model consists of two parts, the **topology** and the **parameters**. In order to support industrial deployment, the model format must be self-complete and must not expose any training source code.
As a result, In PaddlePaddle, the **topology** is represented 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 parameters and efficient serialization/deserialization of parameters.
As a result, In PaddlePaddle, the **topology** is represented 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 parameters and efficient serialization/deserialization of parameters.
## Implementation
## Implementation
The topology is saved as a plain text in a detailed self-contain protobuf file.
The topology is saved as a plain text in a detailed self-contain protobuf file.
The parameters are saved as a binary file. As we all know, the protobuf message has a limit of [64M size](https://developers.google.com/protocol-buffers/docs/reference/cpp/google.protobuf.io.coded_stream#CodedInputStream.SetTotalBytesLimit.details). We have done a [benchmark experiment](https://github.com/PaddlePaddle/Paddle/pull/4610), which shows that protobuf is not fit for the task.
The parameters are saved as a binary file. As we all know, the protobuf message has a limit of [64M size](https://developers.google.com/protocol-buffers/docs/reference/cpp/google.protobuf.io.coded_stream#CodedInputStream.SetTotalBytesLimit.details). We have done a [benchmark experiment](https://github.com/PaddlePaddle/Paddle/pull/4610), which shows that protobuf is not fit for the task.
As a result, we design a particular format for tensor serialization. By default, an 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 all the necessary information, such as the `dims`, and the `LoD` information in [LoDTensor](https://github.com/PaddlePaddle/Paddle/blob/1c0a4c901c9fc881d120249c703b15d1c50dae7d/paddle/framework/lod_tensor.md). A tensor stores values 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,
As a result, we design a particular format for tensor serialization. By default, an 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 all the necessary information, such as the `dims`, and the `LoD` information in [LoDTensor](https://github.com/PaddlePaddle/Paddle/blob/1c0a4c901c9fc881d120249c703b15d1c50dae7d/paddle/framework/lod_tensor.md). A tensor stores values 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,
The table below shows a tensor's byte view in detail. Note that all the signed values are written in the little-endian format.
The table below shows a tensor's byte view in detail. Note that all the signed values are written in the little-endian format.
|field name | type | description |
<table>
| --- | --- | --- |
<thead>
| version | uint32_t | Version of saved file. Always 0 now. |
