mirrors/PaddleClas
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/PaddleClas.git synced 2025-06-03 21:55:06 +08:00
Code Issues Projects Releases Wiki Activity

Merge pull request #160 from WuHaobo/en_docs

Browse Source 
Add quantize en doc.
This commit is contained in:
littletomatodonkey 2020-06-17 19:21:57 +08:00 committed by GitHub
commit 7353212747
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 22 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
docs/en/extension/paddle_mobile_inference.md Normal file
View File

@ -0,0 +1,9 @@
# Paddle-Lite
[Paddle-Lite](https://github.com/PaddlePaddle/Paddle-Lite) is an open-source deep learning framework designed by PaddlePaddle to make it easy to perform inference on mobile, embeded, and IoT devices.
Light Weight is reflected in the use of fewer bits to represent the weight and activation of the neural network,
which can greatly reduce the size of the model,
solve the problem of limited storage space of the terminal device,
and the inference performance is overall better than other frame.
[PaddleClas](https://github.com/PaddlePaddle/PaddleClas) has used Paddle-Lite to evaluate [the performance of the mobile model](../models/Mobile.md).
For more detail of process, please refer to [Paddle-Lite documentations](https://paddle-lite.readthedocs.io/zh/latest/).

13
docs/en/extension/paddle_quantization.md Normal file
View File

@ -0,0 +1,13 @@
# Model Quantifization
Int8 quantization is one of the key features in [PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim).
It supports two kinds of training aware, **Dynamic strategy** and **Static strategy**,
layer-wise and channel-wise quantization,
and using PaddleLite to deploy models generated by PaddleSlim.
By using this toolkit, [PaddleClas](https://github.com/PaddlePaddle/PaddleClas) quantized the mobilenet_v3_large_x1_0 model whose accuracy is 78.9% after distilled.
After quantized, the prediction speed is accelerated from 19.308ms to 14.395ms on SD855.
The storage size is reduced from 21M to 10M.
The top1 recognition accuracy rate is 75.9%.
For specific training methods, please refer to [PaddleSlim quant aware](https://paddlepaddle.github.io/PaddleSlim/quick_start/quant_aware_tutorial.html)。