README.md 2.1 KB
Newer Older
M
MissPenguin 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
# 飞桨训推一体认证(TIPC)开发文档

## 1. TIPC简介

飞桨除了基本的模型训练和预测,还提供了支持多端多平台的高性能推理部署工具。飞桨训推一体认证(TIPC)旨在为用户提供所有飞桨模型的训练推理部署打通情况,即TIPC认证信息,同时提供一套自动化测试工具,方便用户进行一键测试。本文主要介绍了TIPC测试的接入规范。

<div align="center">
    <img src="tipc_guide.png" width="1000">
</div>

## 2. 接入TIPC

对于已完成训练推理功能开发的模型,可以按照本教程接入TIPC测试。模型训练推理功能开发,请参考[飞桨官方文档](https://www.paddlepaddle.org.cn/documentation/docs/zh/guides/index_cn.html)。TIPC接入流程一般分为`开发测试工具``撰写测试文档`两步。

### 2.1 开发测试工具

参考样板间和开发文档完成测试工具开发:

- 测试工具样板间:https://github.com/PaddlePaddle/PaddleOCR/tree/dygraph/test_tipc
- 测试工具开发文档:
B
Bin Lu 已提交
21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37
    - [基础训练预测](./development_specification_docs/train_infer_python.md)
    - [多机多卡训练](./development_specification_docs/fleet_train_infer_python.md)
    - [混合精度训练](./development_specification_docs/amp_train_infer_python.md)
    - [c++预测](./development_specification_docs/inference_cpp.md)
    - [Serving](./development_specification_docs/serving.md)
    - [Lite ARM CPU](./development_specification_docs/Lite_arm_cpu_cpp_infer.md)
    - [Lite OpenCL ARM GPU](./development_specification_docs/Lite_arm_gpu_opencl_cpp_infer.md)
    - [Lite Metal ARM GPU]
    - [Paddle2ONNX](./development_specification_docs/paddle2onnx.md)
    - [Paddle.js]
    - [XPU]
    - [XPU2]
    - [DCU]
    - [NPU]
    - [Jetson](./development_specification_docs/Jeston_infer_python.md)
    - [Windows](./development_specification_docs/Windows_train_infer_python.md)
    - [MacOS](./development_specification_docs/Mac_train_infer_python.md)
M
MissPenguin 已提交
38 39 40 41 42 43

### 2.2 撰写测试文档

参考测试文档样板间撰写测试文档:

- 测试文档样板间:https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/test_tipc/readme.md