From 2136df9750c9fd93c4cdf1a2983f5d5f6636ddf3 Mon Sep 17 00:00:00 2001 From: juncaipeng <52520497+juncaipeng@users.noreply.github.com> Date: Mon, 16 Dec 2019 19:56:25 +0800 Subject: [PATCH] modify model_quantization doc (#2611) * modify model_quantization doc, test=develop --- _all_pages/develop/model_quantization.md | 2 +- _all_pages/v2.1.0/model_quantization.md | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/_all_pages/develop/model_quantization.md b/_all_pages/develop/model_quantization.md index 14daf3581c..6b053e5dad 100644 --- a/_all_pages/develop/model_quantization.md +++ b/_all_pages/develop/model_quantization.md @@ -223,7 +223,7 @@ python compress.py \ 下面以MobileNetV1为例,介绍使用训练后量化方法产出量化模型。关于训练后量化的原理和详细使用方法,请参考[文档](https://github.com/PaddlePaddle/models/tree/develop/PaddleSlim/quant_low_level_api)。 -> 该示例的代码放在[models/PaddleSlim/quant_low_level_api/](https://github.com/PaddlePaddle/models/tree/develop/PaddleSlim/quant_low_level_api)目录下。如果需要执行该示例,首先clone下来[models](https://github.com/PaddlePaddle/models.git),安装具有训练后量化功能的PaddlePaddle,然后执行[run_post_training_quanzation.sh](https://github.com/PaddlePaddle/models/blob/develop/PaddleSlim/quant_low_level_api/run_post_training_quanzation.sh) 脚本,最后量化模型保存在`mobilenetv1_int8_model`目录下。 +> 该示例的代码放在[models/PaddleSlim/quant_low_level_api/](https://github.com/PaddlePaddle/models/tree/develop/PaddleSlim/quant_low_level_api)目录下。如果需要执行该示例,首先clone下来[models](https://github.com/PaddlePaddle/models.git),安装具有训练后量化功能的PaddlePaddle。因为目前Lite支持支持对conv2d、depthwise_conv2d和mul量化,所以修改[run_post_training_quanzation.sh](https://github.com/PaddlePaddle/models/blob/develop/PaddleSlim/quant_low_level_api/run_post_training_quanzation.sh) 脚本,设置is_full_quantize=False,然后执行该脚本;执行结束后,量化模型保存在`mobilenetv1_int8_model`目录下。下面介绍详细步骤。 1)**准备模型和校准数据** diff --git a/_all_pages/v2.1.0/model_quantization.md b/_all_pages/v2.1.0/model_quantization.md index 14daf3581c..6b053e5dad 100644 --- a/_all_pages/v2.1.0/model_quantization.md +++ b/_all_pages/v2.1.0/model_quantization.md @@ -223,7 +223,7 @@ python compress.py \ 下面以MobileNetV1为例,介绍使用训练后量化方法产出量化模型。关于训练后量化的原理和详细使用方法,请参考[文档](https://github.com/PaddlePaddle/models/tree/develop/PaddleSlim/quant_low_level_api)。 -> 该示例的代码放在[models/PaddleSlim/quant_low_level_api/](https://github.com/PaddlePaddle/models/tree/develop/PaddleSlim/quant_low_level_api)目录下。如果需要执行该示例,首先clone下来[models](https://github.com/PaddlePaddle/models.git),安装具有训练后量化功能的PaddlePaddle,然后执行[run_post_training_quanzation.sh](https://github.com/PaddlePaddle/models/blob/develop/PaddleSlim/quant_low_level_api/run_post_training_quanzation.sh) 脚本,最后量化模型保存在`mobilenetv1_int8_model`目录下。 +> 该示例的代码放在[models/PaddleSlim/quant_low_level_api/](https://github.com/PaddlePaddle/models/tree/develop/PaddleSlim/quant_low_level_api)目录下。如果需要执行该示例,首先clone下来[models](https://github.com/PaddlePaddle/models.git),安装具有训练后量化功能的PaddlePaddle。因为目前Lite支持支持对conv2d、depthwise_conv2d和mul量化,所以修改[run_post_training_quanzation.sh](https://github.com/PaddlePaddle/models/blob/develop/PaddleSlim/quant_low_level_api/run_post_training_quanzation.sh) 脚本,设置is_full_quantize=False,然后执行该脚本;执行结束后,量化模型保存在`mobilenetv1_int8_model`目录下。下面介绍详细步骤。 1)**准备模型和校准数据** -- GitLab