mirror of
https://github.com/PaddlePaddle/FastDeploy.git
synced 2025-10-05 16:48:03 +08:00
[Quantization] Update auto compression configs files. (#846)
* Fix links in readme * Fix links in readme * Update PPOCRv2/v3 examples * Update auto compression configs
This commit is contained in:
@@ -11,15 +11,14 @@ FastDeploy提供了一系列高效易用的工具优化部署体验, 提升推
|
||||
FastDeploy基于PaddleSlim的Auto Compression Toolkit(ACT), 给用户提供了一键模型自动化压缩的工具, 用户可以轻松地通过一行命令对模型进行自动化压缩, 并在FastDeploy上部署压缩后的模型, 提升推理速度. 本文档将以FastDeploy一键模型自动化压缩工具为例, 介绍如何安装此工具, 并提供相应的使用文档.
|
||||
|
||||
### 环境准备
|
||||
1.用户参考PaddlePaddle官网, 安装develop版本
|
||||
1.用户参考PaddlePaddle官网, 安装Paddle 2.4 版本
|
||||
```
|
||||
https://www.paddlepaddle.org.cn/install/quick?docurl=/documentation/docs/zh/develop/install/pip/linux-pip.html
|
||||
```
|
||||
|
||||
2.安装PaddleSlim develop版本
|
||||
2.安装PaddleSlim 2.4 版本
|
||||
```bash
|
||||
git clone https://github.com/PaddlePaddle/PaddleSlim.git & cd PaddleSlim
|
||||
python setup.py install
|
||||
pip install paddleslim==2.4.0
|
||||
```
|
||||
|
||||
3.安装fastdeploy-tools工具包
|
||||
|
@@ -11,15 +11,14 @@ FastDeploy provides a series of efficient and easy-to-use tools to optimize the
|
||||
Based on PaddleSlim's Auto Compression Toolkit (ACT), FastDeploy provides users with a one-click model automation compression tool that allows users to easily compress the model with a single command. This document will take FastDeploy's one-click model automation compression tool as an example, introduce how to install the tool, and provide the corresponding documentation for usage.
|
||||
|
||||
### Environmental Preparation
|
||||
1.Install PaddlePaddle develop version
|
||||
1.Install PaddlePaddle 2.4 version
|
||||
```
|
||||
https://www.paddlepaddle.org.cn/install/quick?docurl=/documentation/docs/zh/develop/install/pip/linux-pip.html
|
||||
```
|
||||
|
||||
2.Install PaddleSlim dev version
|
||||
2.Install PaddleSlim 2.4 version
|
||||
```bash
|
||||
git clone https://github.com/PaddlePaddle/PaddleSlim.git & cd PaddleSlim
|
||||
python setup.py install
|
||||
pip install paddleslim==2.4.0
|
||||
```
|
||||
|
||||
3.Install fastdeploy-tools package
|
||||
|
@@ -6,15 +6,14 @@ FastDeploy基于PaddleSlim的Auto Compression Toolkit(ACT), 给用户提供了
|
||||
|
||||
### 环境依赖
|
||||
|
||||
1.用户参考PaddlePaddle官网, 安装develop版本
|
||||
1.用户参考PaddlePaddle官网, 安装Paddle 2.4 版本
|
||||
```
|
||||
https://www.paddlepaddle.org.cn/install/quick?docurl=/documentation/docs/zh/develop/install/pip/linux-pip.html
|
||||
```
|
||||
|
||||
2.安装paddleslim-develop版本
|
||||
2.安装PaddleSlim 2.4 版本
|
||||
```bash
|
||||
git clone https://github.com/PaddlePaddle/PaddleSlim.git & cd PaddleSlim
|
||||
python setup.py install
|
||||
pip install paddleslim==2.4.0
|
||||
```
|
||||
|
||||
### 一键模型自动化压缩工具安装方式
|
||||
|
@@ -7,17 +7,14 @@ We take the Yolov5 series as an example to demonstrate how to install and execut
|
||||
|
||||
### Environment Dependencies
|
||||
|
||||
1. Install the develop version downloaded from PaddlePaddle official website.
|
||||
|
||||
1.Install PaddlePaddle 2.4 version
|
||||
```
|
||||
https://www.paddlepaddle.org.cn/install/quick?docurl=/documentation/docs/zh/develop/install/pip/linux-pip.html
|
||||
```
|
||||
|
||||
2.Install PaddleSlim-develop
|
||||
|
||||
2.Install PaddleSlim 2.4 version
|
||||
```bash
|
||||
git clone https://github.com/PaddlePaddle/PaddleSlim.git & cd PaddleSlim
|
||||
python setup.py install
|
||||
pip install paddleslim==2.4.0
|
||||
```
|
||||
|
||||
### Install Fastdeploy Auto Compression Toolkit
|
||||
|
@@ -24,7 +24,7 @@ Distillation:
|
||||
alpha: 1.0 #蒸馏loss所占权重
|
||||
loss: soft_label #蒸馏loss算法
|
||||
|
||||
Quantization:
|
||||
QuantAware:
|
||||
onnx_format: true #是否采用ONNX量化标准格式, 要在FastDeploy上部署, 必须选true
|
||||
use_pact: true #量化训练是否使用PACT方法
|
||||
activation_quantize_type: 'moving_average_abs_max' #激活量化方式
|
||||
|
@@ -26,7 +26,7 @@ Distillation:
|
||||
alpha: 1.0 #Distillation loss weight
|
||||
loss: soft_label #Distillation loss algorithm
|
||||
|
||||
Quantization:
|
||||
QuantAware:
|
||||
onnx_format: true #Whether to use ONNX quantization standard format or not, must be true to deploy on FastDeploy
|
||||
use_pact: true #Whether to use the PACT method for training
|
||||
activation_quantize_type: 'moving_average_abs_max' #Activations quantization methods
|
||||
|
@@ -17,7 +17,7 @@ Distillation:
|
||||
- softmax_0.tmp_0
|
||||
|
||||
|
||||
Quantization:
|
||||
QuantAware:
|
||||
use_pact: true
|
||||
activation_bits: 8
|
||||
is_full_quantize: false
|
||||
|
@@ -16,7 +16,7 @@ Distillation:
|
||||
node:
|
||||
- softmax_0.tmp_0
|
||||
|
||||
Quantization:
|
||||
QuantAware:
|
||||
use_pact: true
|
||||
activation_bits: 8
|
||||
is_full_quantize: false
|
||||
|
@@ -14,7 +14,7 @@ Distillation:
|
||||
alpha: 1.0
|
||||
loss: soft_label
|
||||
|
||||
Quantization:
|
||||
QuantAware:
|
||||
onnx_format: true
|
||||
use_pact: true
|
||||
activation_quantize_type: 'moving_average_abs_max'
|
||||
|
@@ -14,7 +14,7 @@ Distillation:
|
||||
alpha: 1.0
|
||||
loss: soft_label
|
||||
|
||||
Quantization:
|
||||
QuantAware:
|
||||
onnx_format: true
|
||||
use_pact: true
|
||||
activation_quantize_type: 'moving_average_abs_max'
|
||||
|
@@ -14,7 +14,7 @@ Distillation:
|
||||
alpha: 1.0
|
||||
loss: soft_label
|
||||
|
||||
Quantization:
|
||||
QuantAware:
|
||||
onnx_format: true
|
||||
use_pact: true
|
||||
activation_quantize_type: 'moving_average_abs_max'
|
||||
|
@@ -14,12 +14,13 @@ Distillation:
|
||||
alpha: 1.0
|
||||
loss: soft_label
|
||||
|
||||
Quantization:
|
||||
QuantAware:
|
||||
onnx_format: true
|
||||
activation_quantize_type: 'moving_average_abs_max'
|
||||
quantize_op_types:
|
||||
- conv2d
|
||||
- depthwise_conv2d
|
||||
- conv2d_transpose
|
||||
|
||||
|
||||
PTQ:
|
||||
|
@@ -14,7 +14,7 @@ Distillation:
|
||||
alpha: 1.0
|
||||
loss: soft_label
|
||||
|
||||
Quantization:
|
||||
QuantAware:
|
||||
onnx_format: true
|
||||
activation_quantize_type: 'moving_average_abs_max'
|
||||
quantize_op_types:
|
||||
|
@@ -17,7 +17,7 @@ Distillation:
|
||||
node:
|
||||
- conv2d_94.tmp_0
|
||||
|
||||
Quantization:
|
||||
QuantAware:
|
||||
onnx_format: True
|
||||
quantize_op_types:
|
||||
- conv2d
|
||||
|
Reference in New Issue
Block a user