dockerfile参考，已跑通模型

[he426100]

• 环境
  Ubuntu 22.04.3 LTS
  NVIDIA-SMI 520.61.05 Driver Version: 520.61.05 CUDA Version: 11.8
  3080x2

• 效果
  

• Dockefile

FROM nvidia/cuda:11.8.0-devel-ubuntu22.04

RUN apt-get update && \
    apt-get install -y --no-install-recommends build-essential git wget software-properties-common && \
    add-apt-repository ppa:ondrej/php && apt-get update && \
	DEBIAN_FRONTEND=noninteractive apt-get install -y --no-install-recommends php-cli php-dev && \
	rm -rf /var/lib/apt/lists/*

RUN wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh -O ~/miniconda.sh && \
    chmod +x ~/miniconda.sh && \
    mkdir -p /opt/conda && \
    ~/miniconda.sh -b -u -p /opt/conda && \
    rm ~/miniconda.sh && \
    /opt/conda/bin/conda init bash

ENV PATH="/opt/conda/bin:$PATH"

RUN conda install pytorch torchvision torchaudio pytorch-cuda=11.8 -c pytorch -c nvidia && \
	conda install -c huggingface transformers

RUN git clone https://github.com/swoole/phpy.git /app/phpy

WORKDIR /app/phpy

RUN phpize && \
    ./configure --with-python-dir=/opt/conda && \
    make install && \
    echo "extension=phpy.so" > /etc/php/8.2/cli/conf.d/20_phpy.ini

RUN php -m | grep -i phpy

CMD [ "bash" ]

• 使用

docker run --rm -it --gpus all --name phpy phpy bash
# docker run -d -it --gpus all -v /data/app:/app -v /data/conda:/opt/conda -v /data/cache:/root/.cache --name phpy phpy
cd examples
php pipeline.php

• 安装docker和cuda

sudo apt-get update
sudo apt-get install \
 ca-certificates \
 curl \
 gnupg \
 lsb-release -y
curl -fsSL https://download.docker.com/linux/ubuntu/gpg | sudo gpg --dearmor -o /usr/share/keyrings/docker-archive-keyring.gpg
echo \
"deb [arch=$(dpkg --print-architecture) signed-by=/usr/share/keyrings/docker-archive-keyring.gpg] https://download.docker.com/linux/ubuntu \
$(lsb_release -cs) stable" | sudo tee /etc/apt/sources.list.d/docker.list > /dev/null
sudo apt-get update
sudo apt-get install docker-ce docker-ce-cli containerd.io -y

distribution=$(. /etc/os-release;echo $ID$VERSION_ID) \
&& curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add - \
&& curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list
sudo apt-get update
sudo apt-get install -y nvidia-docker2

wget https://developer.download.nvidia.com/compute/cuda/11.8.0/local_installers/cuda_11.8.0_520.61.05_linux.run 
sudo sh ./cuda_11.8.0_520.61.05_linux.run

[he426100]

• 跑chatglm/cli_demo效果（已实现python版等效的逐字输出效果）
  

• php代码

<?php

define('BASE_PATH', __DIR__ . '/..');
require BASE_PATH . '/vendor/autoload.php';
require __DIR__ . '/utils.php';

use Dotenv\Dotenv;
use Dotenv\Repository\Adapter;
use Dotenv\Repository\RepositoryBuilder;
use function Laravel\Prompts\text;
use function Laravel\Prompts\info;
use function Laravel\Prompts\error;

load_env();

$os = PyCore::import('os');
$platform = PyCore::import('platform');
$transformers = PyCore::import('transformers');
$AutoModel = $transformers->AutoModel;
$AutoTokenizer = $transformers->AutoTokenizer;
$torch = PyCore::import('torch');

$MODEL_PATH = getenv('MODEL_PATH') ?: 'THUDM/chatglm3-6b';
$TOKENIZER_PATH = getenv("TOKENIZER_PATH") ?: $MODEL_PATH;
$DEVICE = $torch->cuda->is_available() ? 'cuda' : 'cpu';

$tokenizer = $AutoTokenizer->from_pretrained($TOKENIZER_PATH, trust_remote_code: true);
if ($DEVICE == 'cuda') {
    # AMD, NVIDIA GPU can use Half Precision
    // $model = $AutoModel->from_pretrained($MODEL_PATH, trust_remote_code: true)->to($DEVICE)->eval();
    $model = load_model_on_gpus($MODEL_PATH, $torch->cuda->device_count());
} else {
    # CPU, Intel GPU and other GPU can use Float16 Precision Only
    $model = $AutoModel->from_pretrained($MODEL_PATH, trust_remote_code: true)->float()->to($DEVICE)->eval();
}

$welcome = '欢迎使用 ChatGLM3-6B 模型，输入内容即可进行对话，clear 清空对话历史，stop 终止程序';

$past_key_values = null;
$history = [];
$stop_stream = false;

info($welcome);

while (true) {
    $query = text('用户：');
    if (trim($query) == 'stop') {
        break;
    } elseif (trim($query) == 'clear') {
        $past_key_values = null;
        $history = [];
        info("\033c");
        info($welcome);
        continue;
    }
    info('ChatGLM: ');
    try {
        $current_length = 0;
        $rs = $model->stream_chat($tokenizer,
            $query,
            history: $history,
            top_p: 1,
            temperature: 0.01,
            past_key_values: $past_key_values,
            return_past_key_values: true
        );
        $it = PyCore::iter($rs);
        echo " \e[32m";
        while ($next = PyCore::next($it)) {
            if ($stop_stream) {
                $stop_stream = false;
                break;
            } else {
                list($response, $history, $past_key_values) = PyCore::scalar($next);
                echo mb_substr($response, $current_length);
                $current_length = mb_strlen($response);
            }
        }
        echo "\e[39m\n";
    } catch (\Throwable $e) {
        error($e->getMessage() ?: '执行出错了');
    }
}

function load_env()
{
	$repository = RepositoryBuilder::createWithNoAdapters()
		->addAdapter(Adapter\PutenvAdapter::class)
		->immutable()
		->make();
	
	Dotenv::create($repository, [BASE_PATH])->safeLoad();
}

[he426100]

运行examples/paddlenlp/test.php教程

• 安装cudnn

1. 从https://developer.nvidia.com/rdp/cudnn-download页面下载cudnn-linux-x86_64-8.9.7.29
2. 从解压后的文件中复制lib和include到cuda目录（按自己环境来，这里是cuda-11.8）

sudo cp -r lib/* /usr/local/cuda-11.8/lib64
sudo cp -r include/* /usr/local/cuda-11.8/include

• 安装飞浆
  https://www.paddlepaddle.org.cn/install/quick?docurl=/documentation/docs/zh/install/pip/linux-pip.html 按这个页面操作就行

• 运行test.php
  

[he426100]

10分钟快速上手飞浆： 手写数字识别任务

• php版代码

<?php

ini_set('memory_limit', '2G');

$paddle = PyCore::import('paddle');
$np = PyCore::import('numpy');
$Normalize = PyCore::import('paddle.vision.transforms')->Normalize;

$transform = $Normalize(mean: [127.5], std: [127.5], data_format: 'CHW');
# 下载数据集并初始化 DataSet
$train_dataset = $paddle->vision->datasets->MNIST(mode: 'train', transform: $transform);
$test_dataset = $paddle->vision->datasets->MNIST(mode: 'test', transform: $transform);

# 模型组网并初始化网络
$lenet = $paddle->vision->models->LeNet(num_classes: 10);
$model = $paddle->Model($lenet);

# 模型训练的配置准备，准备损失函数，优化器和评价指标
$model->prepare(
    $paddle->optimizer->Adam(parameters: $model->parameters()),
    $paddle->nn->CrossEntropyLoss(),
    $paddle->metric->Accuracy()
);

# 模型训练
$model->fit($train_dataset, epochs: 5, batch_size: 64, verbose: 1);
# 模型评估
$model->evaluate($test_dataset, batch_size: 64, verbose: 1);

# 保存模型
$model->save('./output/mnist');
# 加载模型
$model->load('output/mnist');

# 从测试集中取出一张图片
list($img, $label) = $test_dataset->__getitem__(0);
# 将图片shape从1*28*28变为1*1*28*28，增加一个batch维度，以匹配模型输入格式要求
$img_batch = $np->expand_dims($img->astype('float32'), axis: 0);

# 执行推理并打印结果，此处predict_batch返回的是一个list，取出其中数据获得预测结果
$out = $model->predict_batch($img_batch)[0];
$pred_label = $out->argmax();
PyCore::print(PyCore::str('true label: {}, pred label: {}')->format($label->__getitem__(0), $pred_label));

# 可视化图片
$plt = PyCore::import('matplotlib.pyplot');
$plt->imshow($img->__getitem__(0));
# 容器没有gui
$plt->imsave('./output/img.png', $img->__getitem__(0));

• 坑
  转换img, label = test_dataset[0]这一句时有点费劲，试过 list() = PyCore::scalar()，跑不了，不知道有没有更好的办法

• 效果
  

[he426100]

请问是否有仓库镜像，Dockerfile build 出现错误,不知怎么解决 Err:1 https://developer.download.nvidia.cn/compute/cuda/repos/ubuntu2204/x86_64 InRelease The following signatures couldn't be verified because the public key is not available: NO_PUBKEY A4B469963BF863CC Err:3 http://security.ubuntu.com/ubuntu jammy-security InRelease The following signatures couldn't be verified because the public key is not available: NO_PUBKEY 871920D1991BC93C Get:4 http://archive.ubuntu.com/ubuntu jammy-updates InRelease [119 kB] Err:2 http://archive.ubuntu.com/ubuntu jammy InRelease The following signatures couldn't be verified because the public key is not available: NO_PUBKEY 871920D1991BC93C Get:5 http://archive.ubuntu.com/ubuntu jammy-backports InRelease [109 kB] Err:4 http://archive.ubuntu.com/ubuntu jammy-updates InRelease The following signatures couldn't be verified because the public key is not available: NO_PUBKEY 871920D1991BC93C Err:5 http://archive.ubuntu.com/ubuntu jammy-backports InRelease The following signatures couldn't be verified because the public key is not available: NO_PUBKEY 871920D1991BC93C Reading package lists...

你可以基于 modelscope （推荐）、huggingface 、飞浆 等官方镜像安装php环境及phpy扩展。

• 安装php脚本（来自docker-php）

#!/bin/bash

# dependencies required for running "phpize"
# (see persistent deps below)
PHPIZE_DEPS=(autoconf dpkg-dev file g++ gcc libc-dev make pkg-config re2c)

# persistent / runtime deps
set -eux; \
	apt-get update; \
	apt-get install -y --no-install-recommends \
		$PHPIZE_DEPS \
		ca-certificates \
		curl \
        wget \
		xz-utils;

PHP_INI_DIR=/usr/local/etc/php

mkdir -p "$PHP_INI_DIR/conf.d"; \

PHP_VERSION=8.3.0
PHP_URL="https://www.php.net/distributions/php-8.3.0.tar.gz" 

set -eux; \
	\
	savedAptMark="$(apt-mark showmanual)"; \
	apt-get update; \
	apt-get install -y --no-install-recommends gnupg; \
	\
	mkdir -p /usr/src/php; \
	cd /usr/src; \
	\
	wget -c -O php.tar.gz "$PHP_URL"; \
    tar -zxvf php.tar.gz --strip-components=1 -C php;

set -eux; \
	\
	savedAptMark="$(apt-mark showmanual)"; \
	apt-get update; \
	apt-get install -y --no-install-recommends \
		libcurl4-openssl-dev \
		libonig-dev \
		libreadline-dev \
		libsodium-dev \
		libsqlite3-dev \
		libssl-dev \
		libxml2-dev \
		zlib1g-dev \
	; \
	cd /usr/src/php; \
	gnuArch="$(dpkg-architecture --query DEB_BUILD_GNU_TYPE)"; \
	debMultiarch="$(dpkg-architecture --query DEB_BUILD_MULTIARCH)"; \
	if [ ! -d /usr/include/curl ]; then \
		ln -sT "/usr/include/$debMultiarch/curl" /usr/local/include/curl; \
	fi; \
	./configure \
		--build="$gnuArch" \
		--with-config-file-path="$PHP_INI_DIR" \
		--with-config-file-scan-dir="$PHP_INI_DIR/conf.d" \
		\
		\
		--with-mhash \
		\
		--with-pic \
		\
		--enable-mbstring \
		--enable-mysqlnd \
		--with-sodium=shared \
		--with-pdo-sqlite=/usr \
		--with-sqlite3=/usr \
		\
		--with-curl \
		--with-iconv \
		--with-openssl \
		--with-readline \
		--with-zlib \
		\
		--enable-phpdbg \
		--enable-phpdbg-readline \
		\
		--with-pear \
		\
		$(test "$gnuArch" = 's390x-linux-gnu' && echo '--without-pcre-jit') \
		--with-libdir="lib/$debMultiarch" \
		\
		--enable-embed \
                --enable-dom \
                --enable-xml \
                --enable-xmlreader \
                --enable-xmlwriter \
                --enable-soap \
	; \
	make -j "$(nproc)"; \
	make install; \
	\
	cp -v php.ini-* "$PHP_INI_DIR/"; \
	\
	cd /; \
	\
	php --version

• 安装phpy（modelscope-cpu版）

git clone https://github.com/swoole/phpy && cd phpy && \
    phpize && \
    ./configure --with-python-dir=/opt/conda && \
    make install && \
    echo "extension=phpy.so" > /usr/local/etc/php/conf.d/20_phpy.ini && \
    php --ri phpy && \
    curl -sfL https://getcomposer.org/installer | php -- --install-dir=/usr/bin --filename=composer && \
    chmod +x /usr/bin/composer && composer --version && \
    composer install && composer test

• 验证环境

php -r "PyCore::print(PyCore::import('modelscope.pipelines')->pipeline('word-segmentation')('今天天气不错，适合 出去游玩'));"

• 2分钟跑通模型推理

<?php
$pipeline = PyCore::import('modelscope.pipelines')->pipeline;
$word_segmentation = $pipeline('word-segmentation', model: 'damo/nlp_structbert_word-segmentation_chinese-base');

$input_str = '今天天气不错，适合出去游玩';
PyCore::print($word_segmentation($input_str));
// {'output': ['今天', '天气', '不错', '，', '适合', '出去', '游玩']}

[hgc357341051]

能弄个cpu的的吗，只跑验证码识别

[he426100]

能弄个cpu的的吗，只跑验证码识别

示例代码

<?php
/**
 * @link https://modelscope.cn/models/damo/cv_convnextTiny_ocr-recognition-general_damo/summary
 */
$pipeline = PyCore::import('modelscope.pipelines')->pipeline;
$Tasks = PyCore::import('modelscope.utils.constant')->Tasks;
$os = PyCore::import('os');
// 模型可以换成 xiaolv/ocr_small
$pipe = $pipeline($Tasks->ocr_recognition, model: 'damo/cv_convnextTiny_ocr-recognition-general_damo');
$file = './captcha.png';
file_put_contents($file, file_get_contents('https://business.swoole.com/page/captcha_register'));
echo '识别结果：' . $pipe($file)['text'][0], PHP_EOL;

环境

pip config set global.index-url https://mirrors.aliyun.com/pypi/simple
pip config set install.trusted-host mirrors.aliyun.com
pip install -U pip
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu
pip install modelscope transformers SentencePiece opencv-python

[matyhtf]

@he426100 示例可以加入到 phpy 的 examples 中

[he426100]

能弄个cpu的的吗，只跑验证码识别

用下面这个，又快又准
#22

[zeroxxmmbm]

docker-php的安装脚本 基于飞桨官方进行内安装运行后报错 容器环境是ubuntu20的
请问遇到这个报错是 什么问题 dpkg是命令是正常的

++ dpkg-architecture --query DEB_BUILD_GNU_TYPE
php_install.sh: line 51: dpkg-architecture: command not found

[he426100]

docker-php的安装脚本 基于飞桨官方进行内安装运行后报错 容器环境是ubuntu20的 请问遇到这个报错是 什么问题 dpkg是命令是正常的

++ dpkg-architecture --query DEB_BUILD_GNU_TYPE php_install.sh: line 51: dpkg-architecture: command not found

paddle-2.6.0 需要先执行
apt install dpkg-dev pkg-config -y

编译phpy命令需改成

phpize && \
    ./configure --with-python-config=/usr/bin/python3.10-config && \
    make install && \
    echo "extension=phpy.so" > /usr/local/etc/php/conf.d/20_phpy.ini && \
    php --ri phpy

[zeroxxmmbm]

docker-php的安装脚本 基于飞桨官方进行内安装运行后报错 容器环境是ubuntu20的
请问遇到这个报错是 什么问题 dpkg是命令是正常的

++ dpkg-architecture --query DEB_BUILD_GNU_TYPE
php_install.sh: line 51: dpkg-architecture: command not found

docker-php的安装脚本 基于飞桨官方进行内安装运行后报错 容器环境是ubuntu20的 请问遇到这个报错是 什么问题 dpkg是命令是正常的
++ dpkg-architecture --query DEB_BUILD_GNU_TYPE php_install.sh: line 51: dpkg-architecture: command not found

paddle-2.6.0 需要先执行 apt install dpkg-dev pkg-config -y

编译phpy命令需改成

phpize && \
    ./configure --with-python-config=/usr/bin/python3.10-config && \
    make install && \
    echo "extension=phpy.so" > /usr/local/etc/php/conf.d/20_phpy.ini && \
    php --ri phpy

嗯已安装dpkg-dev pkg-config 在测试编译谢谢