From 8610838c127f2393aebc27c6b8b325ab7fce957d Mon Sep 17 00:00:00 2001
From: laugh12321 <laugh12321@vip.qq.com>
Date: Tue, 23 Apr 2024 19:21:52 +0800
Subject: [PATCH] Update README

---
 tools/README.en.md | 22 ++++++++++++++++++++++
 tools/README.md    | 22 ++++++++++++++++++++++
 2 files changed, 44 insertions(+)

diff --git a/tools/README.en.md b/tools/README.en.md
index e69de29..fac5724 100644
--- a/tools/README.en.md
+++ b/tools/README.en.md
@@ -0,0 +1,22 @@
+English | [简体中文](README.md)
+
+# PTQ INT8 Quantization
+
+This is a script for fast PTQ (Post Training Quantization) INT8 quantization using TensorRT, supporting both dynamic and static batching.
+
+## Usage
+
+First, configure the model you want to quantize in `calibration.yaml`.
+
+`calibrator.data` is the path to the data used for calibration, and `calibrator.cache` is the location to save the generated calibration files.
+
+> If you choose **dynamic batching**, ensure that the dimensions of **`batch_shape`** match **`shapes.opt`**. If you choose **static batching**, set **`dynamic`** to **`False`**, and ignore **`shapes`**.
+
+After configuring `calibration.yaml`, run the following command to perform quantization:
+
+```bash
+cd tools
+python ptq_calibration.py
+```
+
+The precision and latency after PTQ quantization vary depending on the model. For maximum precision, it is recommended to use QAT quantization.
\ No newline at end of file
diff --git a/tools/README.md b/tools/README.md
index e69de29..d4b4041 100644
--- a/tools/README.md
+++ b/tools/README.md
@@ -0,0 +1,22 @@
+[English](README.en.md) | 简体中文
+
+# PTQ INT8 量化
+
+这是一个使用 TensorRT 进行快速 PTQ（Post Training Quantization）INT8 量化的脚本，支持动态和静态 Batch。
+
+## 使用方法
+
+首先，在 `calibration.yaml` 中配置你要量化的模型。
+
+`calibrator.data` 是用于校准的数据路径，而 `calibrator.cache` 则是保存生成的校准文件的位置。
+
+> 如果你选择 **动态 Batch**，务必确保 **`batch_shape`** 的维度与 **`shapes.opt`** 一致；如果你选择 **静态 Batch**，将 **`dynamic`** 设为 **`False`**，并**忽略 `shapes`**。
+
+配置好 `calibration.yaml` 后，运行以下命令进行量化：
+
+```bash
+cd tools
+python ptq_calibration.py
+```
+
+PTQ 量化后的精度与延时因模型而异，如果追求最高精度，建议使用 QAT 量化。
\ No newline at end of file