提交 5d6d2bc1 编写于 作者: A Abhinav Arora 提交者: GitHub

Fixing typos and grammatical mistakes in Refactorization documents (#4479)

上级 b9336e6f
# Design Doc: Refactorization Overview # Design Doc: Refactorization Overview
The goal of refactorizaiton include: The goals of refactoring include:
1. Make it easy for external contributors to write new elementory computaiton operations. 1. Making it easy for external contributors to write new elementary computation operations.
1. Make the codebase clean and readable. 1. Making the codebase clean and readable.
1. Introduce a new design of computation representation -- a computation graph of operators and variables. 1. Designing a new computation representation -- a computation graph of operators and variables.
1. The graph representation helps implementing auto-scalable and auto fault recoverable distributed computing. 1. Implementing auto-scalability and auto fault recoverable distributed computing with the help of computation graphs.
## Computation Graphs ## Computation Graphs
1. PaddlePaddle represent the computation, training and inference of DL models, by computation graphs. 1. PaddlePaddle represents the computation, training and inference of Deep Learning models, by computation graphs.
1. Please dig into [computation graphs](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/graph.md) for a solid example. 1. Please refer to [computation graphs](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/graph.md) for a concrete example.
1. Users write Python programs to describe the graphs and run it (locally or remotely). 1. Users write Python programs to describe the graphs and run them (locally or remotely).
1. A graph is composed of *variables* and *operators*. 1. A graph is composed of *variables* and *operators*.
1. The description of graphs must be able to be serialized/deserialized, so it 1. The description of graphs must be capable of being serialized/deserialized, so that
1. could to be sent to the cloud for distributed execution, and 1. It can to be sent to the cloud for distributed execution, and
1. be sent to clients for mobile or enterprise deployment. 1. It can be sent to clients for mobile or enterprise deployment.
1. The Python program do 1. The Python program does the following steps
1. *compilation*: runs a Python program to generate a protobuf message representation of the graph and send it to 1. *compilation*: run a Python program to generate a protobuf message representation of the graph and send it to
1. the C++ library `libpaddle.so` for local execution, 1. the C++ library `libpaddle.so` for local execution,
1. the master process of a distributed training job for training, or 1. the master process of a distributed training job for training, or
1. the server process of a Kubernetes serving job for distributed serving. 1. the server process of a Kubernetes serving job for distributed serving.
1. *execution*: according to the protobuf message, constructs instances of class `Variable` and `OperatorBase`, and run them. 1. *execution*: execute the graph by constructing instances of class [`Variable`](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/variable.h#L24) and [`OperatorBase`](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/operator.h#L70), according to the protobuf message.
## Description and Realization ## Description and Realization of Computation Graph
At compile time, the Python program generates protobuf message representation of the graph, or the description of the graph. At compile time, the Python program generates a protobuf message representation of the graph, or the description of the graph.
At runtime, the C++ program realizes the graph and run it. At runtime, the C++ program realizes the graph and runs it.
| | Representation (protobuf messages) | Realization (C++ class objects) | | | Representation (protobuf messages) | Realization (C++ class objects) |
|---|---|---| |---|---|---|
...@@ -42,30 +42,31 @@ At runtime, the C++ program realizes the graph and run it. ...@@ -42,30 +42,31 @@ At runtime, the C++ program realizes the graph and run it.
|Operation|[OpDesc](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/framework.proto#L35)|[Operator](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/operator.h#L64)| |Operation|[OpDesc](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/framework.proto#L35)|[Operator](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/operator.h#L64)|
|Block|BlockDesc|Block| |Block|BlockDesc|Block|
The word *graph* is exchangable with *block* in this document. A graph represent computation steps and local variables as a C++/Java program block, or a pair of { and }. The word *graph* is interchangeable with *block* in this document. A graph represents computation steps and local variables similar to a C++/Java program block, or a pair of parentheses(`{` and `}`).
## Compilation and Execution ## Compilation and Execution
1. Run an applicaton Python program to describe the graph. In particular, 1. Run an application Python program to describe the graph. In particular, the Python application program does the following:
1. create VarDesc to represent local/intermediate variables, 1. Create `VarDesc` to represent local/intermediate variables,
1. create operators and set attributes, 1. Create operators and set attributes,
1. validate attribute values, 1. Validate attribute values,
1. inference the type and the shape of variables, 1. Infer the type and the shape of variables,
1. plan for memory-reuse for variables, 1. Plan memory-reuse for variables,
1. generate backward and optimization part of the Graph. 1. Generate the backward graph
1. possiblly split the graph for distributed training. 1. Optimize the computation graph.
1. Potentially, split the graph for distributed training.
1. The invocation of `train` or `infer` in the application Python program: 1. The invocation of `train` or [`infer`](https://github.com/PaddlePaddle/Paddle/blob/develop/python/paddle/v2/inference.py#L108) methods in the application Python program does the following:
1. create a new Scope instance in the [scope hierarchy](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/scope.md) for each run of a block, 1. Create a new Scope instance in the [scope hierarchy](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/scope.md) for each run of a block,
1. realize local variables defined in the BlockDesc message in the new scope, 1. realize local variables defined in the BlockDesc message in the new scope,
1. a scope is similar to the stack frame in programming languages, 1. a scope is similar to the stack frame in programming languages,
1. create an instance of class `Block`, in which, 1. Create an instance of class `Block`, in which,
1. realize operators in the BlockDesc message, 1. realize operators in the BlockDesc message,
1. run the Block by calling 1. Run the Block by calling
1. `Block::Eval(vector<Variable>* targets)` for forward and backward computations, or 1. `Block::Eval(vector<Variable>* targets)` for forward and backward computations, or
1. `Block::Eval(vector<Operator>* targets)` for optimization. 1. `Block::Eval(vector<Operator>* targets)` for optimization.
...@@ -76,14 +77,14 @@ The word *graph* is exchangable with *block* in this document. A graph represen ...@@ -76,14 +77,14 @@ The word *graph* is exchangable with *block* in this document. A graph represen
Compile Time -> IR -> Runtime Compile Time -> IR -> Runtime
``` ```
### Benefit ### Benefits of IR
- Optimization - Optimization
```text ```text
Compile Time -> IR -> Optimized IR -> Runtime Compile Time -> IR -> Optimized IR -> Runtime
``` ```
- Send automatically partitioned IR to different nodes. - Automatically send partitioned IR to different nodes.
- Automatic data parallel - Automatic Data Parallelism
```text ```text
Compile Time Compile Time
|-> Single GPU IR |-> Single GPU IR
...@@ -92,7 +93,7 @@ Compile Time -> IR -> Runtime ...@@ -92,7 +93,7 @@ Compile Time -> IR -> Runtime
|-> Node-1 (runs trainer-IR-1) |-> Node-1 (runs trainer-IR-1)
|-> Node-2 (runs pserver-IR) |-> Node-2 (runs pserver-IR)
``` ```
- Automatic model parallel (planned for future) - Automatic Model Parallelism (planned for future)
--- ---
...@@ -105,10 +106,10 @@ Compile Time -> IR -> Runtime ...@@ -105,10 +106,10 @@ Compile Time -> IR -> Runtime
# Operator # Operator
![class_diagram](http://api.paddlepaddle.org/graphviz?dot=https://gist.githubusercontent.com/reyoung/53df507f6749762675dff3e7ce53372f/raw/dd598e8f1976f5759f58af5e5ef94738a6b2e661/op.dot) ![class_diagram](http://api.paddlepaddle.org/graphviz?dot=https://gist.githubusercontent.com/reyoung/53df507f6749762675dff3e7ce53372f/raw/dd598e8f1976f5759f58af5e5ef94738a6b2e661/op.dot)
* `Operator` is the fundamental building block as the user interface. * `Operator` is the fundamental building block of the user interface.
* Operator stores input/output variable name, and attributes. * Operator stores input/output variable names, and attributes.
* The `InferShape` interface is used to infer output variable shapes by its input shapes. * The `InferShape` interface is used to infer the shape of the output variable shapes based on the shapes of the input variables.
* Use `Run` to compute `input variables` to `output variables`. * Use `Run` to compute the `output` variables from the `input` variables.
--- ---
...@@ -126,30 +127,30 @@ Compile Time -> IR -> Runtime ...@@ -126,30 +127,30 @@ Compile Time -> IR -> Runtime
# Why separate Kernel and Operator # Why separate Kernel and Operator
* Separate GPU and CPU code. * Separate GPU and CPU code.
* Make Paddle can run without GPU. * Make Paddle capable of running without GPU.
* Make one operator (which is user interface) can contain many implementations. * Make one operator (which is a user interface) and create many implementations.
* Same mul op, different FP16, FP32 Kernel. different MKL, eigen kernel. * For example, same multiplication op can have different implementations kernels such as FP16 kernel, FP32 kernel, MKL, eigen kernel.
--- ---
# Libraries for Kernel development # Libraries for Kernel development
* `Eigen::Tensor` contains basic math and element-wise functions. * `Eigen::Tensor` contains basic math and element-wise functions.
* Note that `Eigen::Tensor` has broadcast implementation. * Note that `Eigen::Tensor` has broadcast implementation.
* Limit number of `tensor.device(dev) = ` in your code. * Limit the number of `tensor.device(dev) = ` in your code.
* `thrust::tranform` and `std::transform`. * `thrust::tranform` and `std::transform`.
* `thrust` has the same API as C++ standard library. Using `transform` can quickly implement a customized elementwise kernel. * `thrust` has the same API as C++ standard library. Using `transform`, one can quickly implement customized elementwise kernels.
* `thrust` has more complex API, like `scan`, `reduce`, `reduce_by_key`. * `thrust` also has more complex APIs, like `scan`, `reduce`, `reduce_by_key`.
* Hand-writing `GPUKernel` and `CPU` code * Hand-writing `GPUKernel` and `CPU` code
* Do not write `.h`. CPU Kernel should be in `.cc`. GPU kernel should be in `.cu`. (`GCC` cannot compile GPU code.) * Do not write in header (`.h`) files. CPU Kernel should be in cpp source (`.cc`) and GPU kernels should be in cuda (`.cu`) files. (GCC cannot compile GPU code.)
--- ---
# Operator Register # Operator Registration
## Why register is necessary? ## Why registration is necessary?
We need a method to build mappings between Op type names and Op classes. We need a method to build mappings between Op type names and Op classes.
## How to do the register? ## How is registration implemented?
Maintain a map, whose key is the type name and value is corresponding Op constructor. Maintaining a map, whose key is the type name and the value is the corresponding Op constructor.
--- ---
# The Registry Map # The Registry Map
...@@ -177,34 +178,34 @@ REGISTER_OP(op_type, op_class, op_maker_class, grad_op_type, grad_op_class) ...@@ -177,34 +178,34 @@ REGISTER_OP(op_type, op_class, op_maker_class, grad_op_type, grad_op_class)
REGISTER_OP_WITHOUT_GRADIENT(op_type, op_class, op_maker_class) REGISTER_OP_WITHOUT_GRADIENT(op_type, op_class, op_maker_class)
``` ```
### `USE` Macros ### USE Macros
make sure the registration process is executed and linked. Make sure the registration process is executed and linked.
--- ---
# Register Process # Registration Process
1. Write Op class, as well as its gradient Op class if there is. 1. Write an Op class and its gradient Op class, if required.
2. Write Op maker class. In the constructor, describe its inputs, outputs, and attributes. 2. Write an Op maker class. In the constructor of this class, describe the inputs, outputs and attributes of the operator.
3. Invoke macro `REGISTER_OP`. The macro will 3. Invoke the macro `REGISTER_OP`. This macro will
1. call maker class to complete `proto` and `checker` 1. Call maker class to complete the `proto` and the `checker`
2. with the completed `proto` and `checker`, build a new key-value pair in the `OpInfoMap` 2. Using the completed `proto` and `checker`, it will add a new key-value pair to the `OpInfoMap`
4. Invoke `USE` macro in where the Op is used to make sure it is linked. 4. Invoke the `USE` macro in which the Op is used, to make sure that it is linked.
--- ---
# Backward Module (1/2) # Backward Module (1/2)
### Create Backward Operator ### Create Backward Operator
- Mapping from forwarding Op to backward Op - Mapping from forward Op to backward Op
![backward](https://gist.githubusercontent.com/dzhwinter/a6fbd4623ee76c459f7f94591fd1abf0/raw/61026ab6e518e66bde66a889bc42557a1fccff33/backward.png) ![backward](https://gist.githubusercontent.com/dzhwinter/a6fbd4623ee76c459f7f94591fd1abf0/raw/61026ab6e518e66bde66a889bc42557a1fccff33/backward.png)
--- ---
# Backward Module (2/2) # Backward Module (2/2)
### Build Backward Network ### Build Backward Network
- **Input** graph of forwarding operators - **Input**: graph of forwarding operators
- **Output** graph of backward operators - **Output**: graph of backward operators
- **corner case in construction** - **Corner cases in construction**
- shared variable => insert `Add` operator - Shared Variables => insert an `Add` operator to combine gradients
- no gradient => insert `fill_zero_grad` operator - No Gradient => insert a `fill_zero_grad` operator
- recursive netOp => call `Backward` recursively - Recursive NetOp => call `Backward` recursively
- RNN Op => recursively call `Backward` on stepnet - RNN Op => recursively call `Backward` on stepnet
...@@ -213,41 +214,41 @@ make sure the registration process is executed and linked. ...@@ -213,41 +214,41 @@ make sure the registration process is executed and linked.
* `Tensor` is an n-dimension array with type. * `Tensor` is an n-dimension array with type.
* Only dims and data pointers are stored in `Tensor`. * Only dims and data pointers are stored in `Tensor`.
* All operators on `Tensor` is written in `Operator` or global functions. * All operations on `Tensor` are written in `Operator` or global functions.
* variable length Tensor design [LoDTensor](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/lod_tensor.md) * Variable length Tensor design [LoDTensor](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/lod_tensor.md)
* `Variable` is the inputs and outputs of an operator. Not just `Tensor`. * `Variable` instances are the inputs and the outputs of an operator. Not just `Tensor`.
* step_scopes in RNN is a variable and not a tensor. * `step_scopes` in RNN is a variable and not a tensor.
* `Scope` is where variables store at. * `Scope` is where variables are stores.
* map<string/*var name */, Variable> * map<string `variable_name`, Variable>
* `Scope` has a hierarchical structure. The local scope can get variable from its parent scope. * `Scope` has a hierarchical structure. The local scope can get variables from its parent scope.
--- ---
# Block (in design) # Block (in design)
## the difference with original RNNOp ## the difference with original RNNOp
- as an operator is more intuitive than `RNNOp`, - As an operator is more intuitive than `RNNOp`,
- offers new interface `Eval(targets)` to deduce the minimal block to `Run`, - Offers a new interface `Eval(targets)` to deduce the minimal block to `Run`,
- fits the compile-time/ runtime separation design. - Fits the compile-time/ runtime separation design paradigm.
- during the compilation, `SymbolTable` stores `VarDesc`s and `OpDesc`s and serialize to a `BlockDesc` - During the compilation, `SymbolTable` stores `VarDesc`s and `OpDesc`s and serialize to a `BlockDesc`
- when graph executes, a Block with `BlockDesc` passed in creates `Op` and `Var` then `Run` - When graph executes, a Block with `BlockDesc` is passed. It then creates `Op` and `Var` instances and then invokes `Run`.
--- ---
# Milestone # Milestone
- take Paddle/books as the main line, the requirement of the models motivates framework refactoring, - Take Paddle/books as the main line, the requirement of the models motivates framework refactoring,
- model migration - Model migration
- framework development gives **priority support** to model migration, for example, - Framework development gives **priority support** to model migration, for example,
- the MNIST demo needs a Python interface, - the MNIST demo needs a Python interface,
- the RNN models require the framework to support `LoDTensor`. - the RNN models require the framework to support `LoDTensor`.
- determine some timelines, - Determine some timelines,
- heavily-relied Ops need to be migrated first, - Frequently used Ops need to be migrated first,
- different models can be migrated parallelly. - Different models can be migrated in parallel.
- improve the framework at the same time - Improve the framework at the same time
- accept imperfection, concentrated on solving the specific problem at the right price. - Accept imperfection, concentrate on solving the specific problem at the right price.
--- ---
# Control the migration quality # Control the migration quality
- compare the performance of migrated models with old ones. - Compare the performance of migrated models with old ones.
- follow google C style - Follow the google C++ style
- build the automatic workflow of generating Python/C++ documentations - Build the automatic workflow of generating Python/C++ documentations.
- the documentation of layers and ops should be written inside the code - The documentation of layers and ops should be written inside the code.
- take the documentation quality into account when doing PR - Take the documentation quality into account when submitting pull requests.
- preview the documentations, read and improve them from users' perspective - Preview the documentations, read and improve them from a user's perspective.
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