Skip to content

Latest commit

 

History

History
166 lines (110 loc) · 9.04 KB

README_en.md

File metadata and controls

166 lines (110 loc) · 9.04 KB

X2Paddle

PyPI - X2Paddle Version PyPI Status License Version python version

Introduction

X2Paddle is a model conversion tool under the PaddlePaddle ecosystem, dedicated to helping users of other deep learning frameworks to quickly migrate to PaddlePaddle framework. Currently supports inference model conversion and PyTorch training code migration. We also provide a detailed API comparison document between different frameworks, to reduce the time and cost of developers to migrate their models to PaddlePaddle.

Features

  • Supports major deep learning frameworks

    • It has supported the conversion on inference models of Caffe/TensorFlow/ONNX/PyTorch and the conversion of PyTorch training projects, which covers the major deep learning frameworks in the market at present. For further details, refer to support model zoo list and test demo

  • Rich set of supported models

    • Supports most model conversion on mainstream CV and NLP models. Currently X2Paddle supports 130+ PyTorch OPs, 90+ ONNX OPs, 90+ TensorFlow OPs and 30+ Caffe OPs. For further details, refer to support list

  • Simple and easy to use

    • Model conversion can be done with a single command or an API.

Capabilities

  • Inference model conversion

    • Support one-stop model conversion from Caffe/TensorFlow/ONNX/PyTorch to PaddlePaddle inference model, and use PaddleInference/PaddleLite for CPU/GPU/Arm and other devices deployment.

  • PyTorch training projects conversion

    • Support one-stop conversion of PyTorch project Python code (including training, prediction) into the project based on PaddlePaddle framework, helping developers quickly migrate their project quickly. And can enjoy AIStudio platform, which provide a large number of free computational power. New feature, try it!

  • API Documentation

    • Detailed comparative analysis of API documentation to help developers quickly migrate from the use of the PyTorch framework to the use of PaddlePaddle framework, greatly reducing the cost of learning. New content, learn about!

Installation

Environment Dependencies

  • python >= 3.8
  • paddlepaddle >= 2.2.2 (Officially verified to 3.0.0beta1)
  • tensorflow == 1.14 (For TensorFlow model conversion. Where test_benchmark models have been tested on 2.16.1.)
  • onnx >= 1.6.0 (For ONNX model conversion. Where test_benchmark models have been tested on 1.17.0.)
  • torch >= 1.5.0 (For PyTorch model conversion. Where test_benchmark models have been tested on 2.4.1.)
  • paddlelite >= 2.9.0 (For conversion to Paddle-Lite supported formats, the latest version is recommended.)

Note: The above tested versions do not mean that X2Paddle supports all the operators in the corresponding version, it only means that they can be used in this environment. Please refer to support List

pip Installation (Recommended)

For a stable version, install X2Paddle via pip:

pip install x2paddle

Source Code Installation

If you want to experience the latest features, you can use the source code installation method:

git clone https://github.com/PaddlePaddle/X2Paddle.git
cd X2Paddle
git checkout develop
python setup.py install

Quick Start

Feature 1: Inference Model Conversion

PyTorch model Conversion

from x2paddle.convert import pytorch2paddle
pytorch2paddle(module=torch_module,
               save_dir="./pd_model",
               jit_type="trace",
               input_examples=[torch_input])
# module (torch.nn.Module): PyTorch's Module
# save_dir (str): Save path of inference model
# jit_type (str): Convert approach. Default is "trace"
# input_examples (list[torch.tensor]): torch.nn.Module's input. The length of list must be equal to the input's. Default is None.

The script approach more details can be found PyTorch model conversion documentation.

TensorFlow Model Conversion

x2paddle --framework=tensorflow --model=tf_model.pb --save_dir=pd_model

ONNX Model Conversion

x2paddle --framework=onnx --model=onnx_model.onnx --save_dir=pd_model

Caffe Model Conversion

x2paddle --framework=caffe --prototxt=deploy.prototxt --weight=deploy.caffemodel --save_dir=pd_model

Conversion Parameter Description

Parameters Description
--framework Source model type (TensorFlow, Caffe, ONNX)
--prototxt When the framework is Caffe, this parameter specifies the path to the proto file of the caffe model
--weight When the framework is Caffe, this parameter specifies the path to the parameter file for the caffe model
--save_dir Specify the path to the directory where the converted model is saved
--model When framework is TensorFlow/ONNX, this parameter specifies TensorFlow's pb file path or ONNX's model file path
--input_shape_dict [optional] For ONNX, ONNX input shape
--caffe_proto [optional] Path to the caffe_pb2.py file compiled from caffe.proto, used when a custom Layer exists, default is None
--define_input_shape [optional] For TensorFlow, When this parameter is specified, forces the user to enter the shape of each Placeholder, see documentation Q2
--enable_code_optim [optional] For PyTorch, Whether to optimize the generated code, default is False
--to_lite [optional] Whether to use the opt tool to convert to Paddle-Lite supported formats, default is False
--lite_valid_places [optional] Specify the type of conversion, you can specify more than one backend at the same time (separated by commas), opt will automatically select the best way, default is arm.
--lite_model_type [optional] Specify the model conversion type, currently supports two types: protobuf and naive_buffer, default is naive_buffer
--disable_feedback [optional] Whether or not to turn off X2Paddle feedback; By default, X2Paddle will count the success rate in model conversion, as well as the source of the conversion framework and other information, in order to help X2Paddle iterate according to the user's needs. X2Paddle will not upload the user's model files. If you don't want to participate in the feedback, you can specify this parameter as False.

X2Paddle API

Currently X2Paddle provides API to convert models, you can refer to X2PaddleAPI

One-Stop Conversion of Paddle-Lite Supported Formats

Refer to convert2lite_api

Feature 2: PyTorch Project Migration

Project conversion consists of 3 steps

  1. project code preprocessing
  2. one-stop code/pre-trained model conversion
  3. post-processing of the converted code

Refer to pytorch_project_convertor

Model Conversion with VisualDL

VisualDL, the PaddlePaddle visualization tool, has deployed the model conversion tool on the official website to provide service, you can click Service Link to perform online ONNX2Paddle model conversion.

ONNX2Paddle

Tutorials

  1. TensorFlow Inference Model Conversion Tutorial
  2. MMDetection Model Conversion Guide
  3. PyTorch Inference Model Conversion Tutorial
  4. PyTorch Training Project Conversion Tutorial

🤗Contributing🤗

You are more than welcome to contribute code to X2Paddle or provide suggestions on how to use it.