M1 Mac os(Apple Silicon)天生一对Python3开发环境搭建(集成深度学习框架Tensorflow/Pytorch)

时间：10-27来源：作者：点击数：87

笔者投入M1的怀抱已经有一段时间了，俗话说得好，但闻新人笑，不见旧人哭，Intel mac早已被束之高阁，而M1 mac已经不能用真香来形容了，简直就是“香透满堂金玉彩，扇遮半面桃花开！”，轻抚M1 mac那滑若柔荑的秒控键盘，别说996了，就是007，我们也能安之若素，也可以笑慰平生。好了，日常吹M1的环节结束，正所谓剑虽利,不厉不断，材虽美，不学不高。本次我们尝试在M1 Mac os 中搭建Python3的开发环境。

一般情况下，直接Python官网(python.org)下载最新的基于arm架构的python3.9即可，但是由于向下兼容等问题，我们尝试使用Python多版本管理软件conda，conda在业界有三大巨头分别是:Anaconda、Miniconda以及Condaforge，虽然都放出消息要适配M1芯片，但是目前最先放出稳定版的是Condaforge，进入下载页面：https://github.com/conda-forge/miniforge/#download 选择mac arm64位架构:

该文件并不是安装包，而是一个shell脚本，下载成功后，进入命令行目录:

cd ~/Downloads

执行命令进行安装：

sudo bash ./Miniforge3-MacOSX-arm64.sh

随后会有一些条款需要确认，这里按回车之后键入yes:

Welcome to Miniforge3 4.9.2-7

In order to continue the installation process, please review the license
agreement.
Please, press ENTER to continue
>>> 
BSD 3-clause license
Copyright (c) 2019-2020, conda-forge
All rights reserved.

Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are met:

1. Redistributions of source code must retain the above copyright notice, this
list of conditions and the following disclaimer.

2. Redistributions in binary form must reproduce the above copyright notice,
this list of conditions and the following disclaimer in the documentation
and/or other materials provided with the distribution.

3. Neither the name of the copyright holder nor the names of its contributors
may be used to endorse or promote products derived from this software without
specific prior written permission.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.


Do you accept the license terms? [yes|no]
[no] >>> yes

安装的默认版本还是3.9，会附带安装35个基础库，这样就不用我们自己手动安装了：

brotlipy                    0.7.0  py39h46acfd9_1001   installed         
  bzip2                       1.0.8  h27ca646_4          installed         
  ca-certificates         2020.12.5  h4653dfc_0          installed         
  certifi                 2020.12.5  py39h2804cbe_1      installed         
  cffi                       1.14.5  py39h702c04f_0      installed         
  chardet                     4.0.0  py39h2804cbe_1      installed         
  conda                       4.9.2  py39h2804cbe_0      installed         
  conda-package-handling      1.7.2  py39h51e6412_0      installed         
  cryptography                3.4.4  py39h6e07874_0      installed         
  idna                         2.10  pyh9f0ad1d_0        installed         
  libcxx                     11.0.1  h168391b_0          installed         
  libffi                        3.3  h9f76cd9_2          installed         
  ncurses                       6.2  h9aa5885_4          installed         
  openssl                    1.1.1j  h27ca646_0          installed         
  pip                        21.0.1  pyhd8ed1ab_0        installed         
  pycosat                     0.6.3  py39h46acfd9_1006   installed         
  pycparser                    2.20  pyh9f0ad1d_2        installed         
  pyopenssl                  20.0.1  pyhd8ed1ab_0        installed         
  pysocks                     1.7.1  py39h2804cbe_3      installed         
  python                      3.9.2  hcbd9b3a_0_cpython  installed         
  python_abi                    3.9  1_cp39              installed         
  readline                      8.0  hc8eb9b7_2          installed         
  requests                   2.25.1  pyhd3deb0d_0        installed         
  ruamel_yaml               0.15.80  py39h46acfd9_1004   installed         
  setuptools                 49.6.0  py39h2804cbe_3      installed         
  six                        1.15.0  pyh9f0ad1d_0        installed         
  sqlite                     3.34.0  h6d56c25_0          installed         
  tk                         8.6.10  hf7e6567_1          installed         
  tqdm                       4.57.0  pyhd8ed1ab_0        installed         
  tzdata                      2021a  he74cb21_0          installed         
  urllib3                    1.26.3  pyhd8ed1ab_0        installed         
  wheel                      0.36.2  pyhd3deb0d_0        installed         
  xz                          5.2.5  h642e427_1          installed         
  yaml                        0.2.5  h642e427_0          installed         
  zlib                       1.2.11  h31e879b_1009       installed

然后编辑配置文件vim ~/.zshrc，加入如下内容(此处liuyue是笔者用户名，需改成你的Mac当前用户名)：

path=('/Users/liuyue/miniforge3/bin' $path)
export PATH

存盘之后执行命令：

source ~/.zshrc

配置好环境变量之后，键入python3：

➜  ~ python3
Python 3.9.2 | packaged by conda-forge | (default, Feb 21 2021, 05:00:30) 
[Clang 11.0.1 ] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>>

可以看到已经使用conda安装的python版本了。

这里简单介绍一下conda命令：

conda info 可以查看当前conda的基本信息内核，平台，下载源以及目录位置：

➜  ~ conda info

     active environment : None
       user config file : /Users/liuyue/.condarc
 populated config files : /Users/liuyue/miniforge3/.condarc
          conda version : 4.9.2
    conda-build version : not installed
         python version : 3.9.2.final.0
       virtual packages : __osx=11.2.2=0
                          __unix=0=0
                          __archspec=1=arm64
       base environment : /Users/liuyue/miniforge3  (read only)
           channel URLs : https://conda.anaconda.org/conda-forge/osx-arm64
                          https://conda.anaconda.org/conda-forge/noarch
          package cache : /Users/liuyue/miniforge3/pkgs
                          /Users/liuyue/.conda/pkgs
       envs directories : /Users/liuyue/.conda/envs
                          /Users/liuyue/miniforge3/envs
               platform : osx-arm64
             user-agent : conda/4.9.2 requests/2.25.1 CPython/3.9.2 Darwin/20.3.0 OSX/11.2.2
                UID:GID : 502:20
             netrc file : None
           offline mode : False

由于一些众所周知的学术问题，我们需要配置一下国内下载源：

conda config --add channels https://mirrors.ustc.edu.cn/anaconda/pkgs/main/
conda config --add channels https://mirrors.ustc.edu.cn/anaconda/pkgs/free/
conda config --add channels https://mirrors.ustc.edu.cn/anaconda/cloud/conda-forge/
conda config --add channels https://mirrors.ustc.edu.cn/anaconda/cloud/msys2/
conda config --add channels https://mirrors.ustc.edu.cn/anaconda/cloud/bioconda/
conda config --add channels https://mirrors.ustc.edu.cn/anaconda/cloud/menpo/

conda config --set show_channel_urls yes

随后查看当前下载源：

conda config --show

可以看到国内源已经被添加进去了：

channel_priority: flexible
channels:
  - https://mirrors.ustc.edu.cn/anaconda/cloud/menpo/
  - https://mirrors.ustc.edu.cn/anaconda/cloud/bioconda/
  - https://mirrors.ustc.edu.cn/anaconda/cloud/msys2/
  - https://mirrors.ustc.edu.cn/anaconda/cloud/conda-forge/
  - https://mirrors.ustc.edu.cn/anaconda/pkgs/free/
  - https://mirrors.ustc.edu.cn/anaconda/pkgs/main/
  - defaults
  - conda-forge
client_ssl_cert: None

其他的一些conda常用命令：

1. conda --version #查看conda版本，验证是否安装

2. conda update conda #更新至最新版本，也会更新其它相关包

3. conda update --all #更新所有包

4. conda update package_name #更新指定的包

5. conda create -n env_name package_name #创建名为env_name的新环境，并在该环境下安装名为package_name 的包，可以指定新环境的版本号，例如：conda create -n python2 python=python2.7 numpy pandas，创建了python2环境，python版本为2.7，同时还安装了numpy pandas包

6. source activate env_name #切换至env_name环境

7. source deactivate #退出环境

8. conda info -e #显示所有已经创建的环境

9. conda create --name new_env_name --clone old_env_name #复制old_env_name为new_env_name

10. conda remove --name env_name –all #删除环境

11. conda list #查看所有已经安装的包

12. conda install package_name #在当前环境中安装包

13. conda install --name env_name package_name #在指定环境中安装包

14. conda remove -- name env_name package #删除指定环境中的包

15. conda remove package #删除当前环境中的包

16. conda create -n tensorflow_env tensorflow

conda activate tensorflow_env #conda 安装tensorflow的CPU版本

17. conda create -n tensorflow_gpuenv tensorflow-gpu

conda activate tensorflow_gpuenv #conda安装tensorflow的GPU版本

18. conda env remove -n env_name #采用第10条的方法删除环境失败时，可采用这种方法

接着我们来尝试集成深度学习框架Tensorflow，由于目前默认是3.9，我们使用conda创建一个3.8的虚拟开发环境：

sudo conda create -n py38 python=3.8

安装成功后，输入命令：

conda info -e

就可以查看当前conda安装的所有版本：

➜  ~ conda info -e
# conda environments:
#
base                  *  /Users/liuyue/miniforge3
py38                     /Users/liuyue/miniforge3/envs/py38

可以看到一个默认的3.9环境，和新装的3.8环境，星号代表当前所处的环境，这里我们切换到3.8：

conda activate py38

此时环境已经切换到3.8：

(py38) ➜  ~ conda activate py38 
(py38) ➜  ~ conda info -e
# conda environments:
#
base                     /Users/liuyue/miniforge3
py38                  *  /Users/liuyue/miniforge3/envs/py38

(py38) ➜  ~

下面开启深度学习框架Tensorflow之旅，由于苹果对m1芯片单独做了适配，所以不能用以前的pip方式直接进行安装，需要单独下载文件：https://github.com/apple/tensorflow_macos/releases/download/v0.1alpha1/tensorflow_macos-0.1alpha1.tar.gz

解压文件：

tar -zxvf tensorflow_macos-0.1alpha1.tar.gz

解压后进入目录(一定要进入arm64的文件内)：

cd tensorflow_macos/arm64

执行命令利用下载的arm64内核安装包进行安装：

pip install --force pip==20.2.4 wheel setuptools cached-property six

pip install --upgrade --no-dependencies --force numpy-1.18.5-cp38-cp38-macosx_11_0_arm64.whl grpcio-1.33.2-cp38-cp38-macosx_11_0_arm64.whl h5py-2.10.0-cp38-cp38-macosx_11_0_arm64.whl tensorflow_addons-0.11.2+mlcompute-cp38-cp38-macosx_11_0_arm64.whl

pip install absl-py astunparse flatbuffers gast google_pasta keras_preprocessing opt_einsum protobuf tensorflow_estimator termcolor typing_extensions wrapt wheel tensorboard typeguard

pip install --upgrade --force --no-dependencies tensorflow_macos-0.1a1-cp38-cp38-macosx_11_0_arm64.whl

安装成功后，测试一下：

(py38) ➜  arm64 python
Python 3.8.8 | packaged by conda-forge | (default, Feb 20 2021, 15:50:57) 
[Clang 11.0.1 ] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> import tensorflow
>>>

没有任何问题。

下面我们来测试一下M1通过Tensorflow训练模型的效率，还记得衣香鬓影的“机械姬”吗：人工智能不过尔尔，基于Python3深度学习库Keras/TensorFlow打造属于自己的聊天机器人(ChatRobot)

编写my_chat.py:

import nltk
import ssl
from nltk.stem.lancaster import LancasterStemmer
stemmer = LancasterStemmer()

import numpy as np
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Activation, Dropout
from tensorflow.keras.optimizers import SGD
import pandas as pd
import pickle
import random


words = []
classes = []
documents = []
ignore_words = ['?']
# loop through each sentence in our intents patterns


intents = {"intents": [
        {"tag": "打招呼",
         "patterns": ["你好", "您好", "请问", "有人吗", "师傅","不好意思","美女","帅哥","靓妹"],
         "responses": ["您好", "又是您啊", "吃了么您内","您有事吗"],
         "context": [""]
        },
        {"tag": "告别",
         "patterns": ["再见", "拜拜", "88", "回见", "回头见"],
         "responses": ["再见", "一路顺风", "下次见", "拜拜了您内"],
         "context": [""]
        },
   ]
}

for intent in intents['intents']:
    for pattern in intent['patterns']:
        # tokenize each word in the sentence
        w = nltk.word_tokenize(pattern)
        # add to our words list
        words.extend(w)
        # add to documents in our corpus
        documents.append((w, intent['tag']))
        # add to our classes list
        if intent['tag'] not in classes:
            classes.append(intent['tag'])
# stem and lower each word and remove duplicates
words = [stemmer.stem(w.lower()) for w in words if w not in ignore_words]
words = sorted(list(set(words)))
# sort classes
classes = sorted(list(set(classes)))
# documents = combination between patterns and intents
# print (len(documents), "documents")
# # classes = intents
# print (len(classes), "语境", classes)
# # words = all words, vocabulary
# print (len(words), "词数", words)


# create our training data
training = []
# create an empty array for our output
output_empty = [0] * len(classes)
# training set, bag of words for each sentence
for doc in documents:
    # initialize our bag of words
    bag = []

    pattern_words = doc[0]
   
    pattern_words = [stemmer.stem(word.lower()) for word in pattern_words]

    for w in words:
        bag.append(1) if w in pattern_words else bag.append(0)
    
 
    output_row = list(output_empty)
    output_row[classes.index(doc[1])] = 1
    
    training.append([bag, output_row])

random.shuffle(training)
training = np.array(training)

train_x = list(training[:,0])
train_y = list(training[:,1])


model = Sequential()
model.add(Dense(128, input_shape=(len(train_x[0]),), activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(64, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(len(train_y[0]), activation='softmax'))


sgd = SGD(lr=0.01, decay=1e-6, momentum=0.9, nesterov=True)
model.compile(loss='categorical_crossentropy', optimizer=sgd, metrics=['accuracy'])


model.fit(np.array(train_x), np.array(train_y), epochs=200, batch_size=5, verbose=1)



def clean_up_sentence(sentence):
    # tokenize the pattern - split words into array
    sentence_words = nltk.word_tokenize(sentence)
    # stem each word - create short form for word
    sentence_words = [stemmer.stem(word.lower()) for word in sentence_words]
    return sentence_words
# return bag of words array: 0 or 1 for each word in the bag that exists in the sentence
def bow(sentence, words, show_details=True):
    # tokenize the pattern
    sentence_words = clean_up_sentence(sentence)
    # bag of words - matrix of N words, vocabulary matrix
    bag = [0]*len(words)  
    for s in sentence_words:
        for i,w in enumerate(words):
            if w == s: 
                # assign 1 if current word is in the vocabulary position
                bag[i] = 1
                if show_details:
                    print ("found in bag: %s" % w)
    return(np.array(bag))


def classify_local(sentence):
    ERROR_THRESHOLD = 0.25
    
    # generate probabilities from the model
    input_data = pd.DataFrame([bow(sentence, words)], dtype=float, index=['input'])
    results = model.predict([input_data])[0]
    # filter out predictions below a threshold, and provide intent index
    results = [[i,r] for i,r in enumerate(results) if r>ERROR_THRESHOLD]
    # sort by strength of probability
    results.sort(key=lambda x: x[1], reverse=True)
    return_list = []
    for r in results:
        return_list.append((classes[r[0]], str(r[1])))
    # return tuple of intent and probability
    
    return return_list


p = bow("你好", words)
print (p)

print(classify_local('请问'))

返回结果：

(py38) ➜  mytornado git:(master) ✗ python3 test_mychat.py
2021-03-03 22:43:21.059383: I tensorflow/compiler/mlir/mlir_graph_optimization_pass.cc:116] None of the MLIR optimization passes are enabled (registered 2)
2021-03-03 22:43:21.059529: W tensorflow/core/platform/profile_utils/cpu_utils.cc:126] Failed to get CPU frequency: 0 Hz
Epoch 1/200
3/3 [==============================] - 0s 570us/step - loss: 0.6966 - accuracy: 0.5750
Epoch 2/200
3/3 [==============================] - 0s 482us/step - loss: 0.6913 - accuracy: 0.4857
Epoch 3/200
3/3 [==============================] - 0s 454us/step - loss: 0.6795 - accuracy: 0.4750
Epoch 4/200
3/3 [==============================] - 0s 434us/step - loss: 0.6913 - accuracy: 0.4750
Epoch 5/200
3/3 [==============================] - 0s 417us/step - loss: 0.6563 - accuracy: 0.5107
Epoch 6/200
3/3 [==============================] - 0s 454us/step - loss: 0.6775 - accuracy: 0.5714
Epoch 7/200
3/3 [==============================] - 0s 418us/step - loss: 0.6582 - accuracy: 0.6964
Epoch 8/200
3/3 [==============================] - 0s 487us/step - loss: 0.6418 - accuracy: 0.8071
Epoch 9/200
3/3 [==============================] - 0s 504us/step - loss: 0.6055 - accuracy: 0.6964
Epoch 10/200
3/3 [==============================] - 0s 457us/step - loss: 0.5933 - accuracy: 0.6964
Epoch 11/200
3/3 [==============================] - 0s 392us/step - loss: 0.6679 - accuracy: 0.5714
Epoch 12/200
3/3 [==============================] - 0s 427us/step - loss: 0.6060 - accuracy: 0.7464
Epoch 13/200
3/3 [==============================] - 0s 425us/step - loss: 0.6677 - accuracy: 0.5964
Epoch 14/200
3/3 [==============================] - 0s 420us/step - loss: 0.6208 - accuracy: 0.6214
Epoch 15/200
3/3 [==============================] - 0s 401us/step - loss: 0.6315 - accuracy: 0.6714
Epoch 16/200
3/3 [==============================] - 0s 401us/step - loss: 0.6718 - accuracy: 0.6464
Epoch 17/200
3/3 [==============================] - 0s 386us/step - loss: 0.6407 - accuracy: 0.6714
Epoch 18/200
3/3 [==============================] - 0s 505us/step - loss: 0.6031 - accuracy: 0.6464
Epoch 19/200
3/3 [==============================] - 0s 407us/step - loss: 0.6245 - accuracy: 0.6214
Epoch 20/200
3/3 [==============================] - 0s 422us/step - loss: 0.5805 - accuracy: 0.6964
Epoch 21/200
3/3 [==============================] - 0s 379us/step - loss: 0.6923 - accuracy: 0.5464
Epoch 22/200
3/3 [==============================] - 0s 396us/step - loss: 0.6383 - accuracy: 0.5714
Epoch 23/200
3/3 [==============================] - 0s 427us/step - loss: 0.6628 - accuracy: 0.5714
Epoch 24/200
3/3 [==============================] - 0s 579us/step - loss: 0.6361 - accuracy: 0.5964
Epoch 25/200
3/3 [==============================] - 0s 378us/step - loss: 0.5632 - accuracy: 0.7214
Epoch 26/200
3/3 [==============================] - 0s 387us/step - loss: 0.6851 - accuracy: 0.5214
Epoch 27/200
3/3 [==============================] - 0s 393us/step - loss: 0.6012 - accuracy: 0.6214
Epoch 28/200
3/3 [==============================] - 0s 392us/step - loss: 0.6470 - accuracy: 0.5964
Epoch 29/200
3/3 [==============================] - 0s 348us/step - loss: 0.6346 - accuracy: 0.6214
Epoch 30/200
3/3 [==============================] - 0s 362us/step - loss: 0.6350 - accuracy: 0.4964
Epoch 31/200
3/3 [==============================] - 0s 369us/step - loss: 0.5842 - accuracy: 0.5964
Epoch 32/200
3/3 [==============================] - 0s 481us/step - loss: 0.5279 - accuracy: 0.7214
Epoch 33/200
3/3 [==============================] - 0s 439us/step - loss: 0.5956 - accuracy: 0.7321
Epoch 34/200
3/3 [==============================] - 0s 355us/step - loss: 0.5570 - accuracy: 0.6964
Epoch 35/200
3/3 [==============================] - 0s 385us/step - loss: 0.5546 - accuracy: 0.8071
Epoch 36/200
3/3 [==============================] - 0s 375us/step - loss: 0.5616 - accuracy: 0.6714
Epoch 37/200
3/3 [==============================] - 0s 379us/step - loss: 0.6955 - accuracy: 0.6464
Epoch 38/200
3/3 [==============================] - 0s 389us/step - loss: 0.6089 - accuracy: 0.7321
Epoch 39/200
3/3 [==============================] - 0s 375us/step - loss: 0.5377 - accuracy: 0.6714
Epoch 40/200
3/3 [==============================] - 0s 392us/step - loss: 0.6224 - accuracy: 0.7179
Epoch 41/200
3/3 [==============================] - 0s 379us/step - loss: 0.6234 - accuracy: 0.5464
Epoch 42/200
3/3 [==============================] - 0s 411us/step - loss: 0.5224 - accuracy: 0.8321
Epoch 43/200
3/3 [==============================] - 0s 386us/step - loss: 0.5848 - accuracy: 0.5964
Epoch 44/200
3/3 [==============================] - 0s 401us/step - loss: 0.4620 - accuracy: 0.8679
Epoch 45/200
3/3 [==============================] - 0s 365us/step - loss: 0.4664 - accuracy: 0.8071
Epoch 46/200
3/3 [==============================] - 0s 367us/step - loss: 0.5904 - accuracy: 0.7679
Epoch 47/200
3/3 [==============================] - 0s 359us/step - loss: 0.5111 - accuracy: 0.7929
Epoch 48/200
3/3 [==============================] - 0s 363us/step - loss: 0.4712 - accuracy: 0.8679
Epoch 49/200
3/3 [==============================] - 0s 401us/step - loss: 0.5601 - accuracy: 0.8071
Epoch 50/200
3/3 [==============================] - 0s 429us/step - loss: 0.4884 - accuracy: 0.7929
Epoch 51/200
3/3 [==============================] - 0s 377us/step - loss: 0.5137 - accuracy: 0.8286
Epoch 52/200
3/3 [==============================] - 0s 368us/step - loss: 0.5475 - accuracy: 0.8286
Epoch 53/200
3/3 [==============================] - 0s 592us/step - loss: 0.4077 - accuracy: 0.8536
Epoch 54/200
3/3 [==============================] - 0s 400us/step - loss: 0.5367 - accuracy: 0.8179
Epoch 55/200
3/3 [==============================] - 0s 399us/step - loss: 0.5288 - accuracy: 0.8429
Epoch 56/200
3/3 [==============================] - 0s 367us/step - loss: 0.5775 - accuracy: 0.6964
Epoch 57/200
3/3 [==============================] - 0s 372us/step - loss: 0.5680 - accuracy: 0.6821
Epoch 58/200
3/3 [==============================] - 0s 360us/step - loss: 0.5164 - accuracy: 0.7321
Epoch 59/200
3/3 [==============================] - 0s 364us/step - loss: 0.5334 - accuracy: 0.6571
Epoch 60/200
3/3 [==============================] - 0s 358us/step - loss: 0.3858 - accuracy: 0.9036
Epoch 61/200
3/3 [==============================] - 0s 356us/step - loss: 0.4313 - accuracy: 0.8679
Epoch 62/200
3/3 [==============================] - 0s 373us/step - loss: 0.5017 - accuracy: 0.8429
Epoch 63/200
3/3 [==============================] - 0s 346us/step - loss: 0.4649 - accuracy: 0.8429
Epoch 64/200
3/3 [==============================] - 0s 397us/step - loss: 0.3804 - accuracy: 0.8893
Epoch 65/200
3/3 [==============================] - 0s 361us/step - loss: 0.5030 - accuracy: 0.7929
Epoch 66/200
3/3 [==============================] - 0s 372us/step - loss: 0.3958 - accuracy: 0.9286
Epoch 67/200
3/3 [==============================] - 0s 345us/step - loss: 0.4240 - accuracy: 0.8536
Epoch 68/200
3/3 [==============================] - 0s 360us/step - loss: 0.4651 - accuracy: 0.7929
Epoch 69/200
3/3 [==============================] - 0s 376us/step - loss: 0.4687 - accuracy: 0.7571
Epoch 70/200
3/3 [==============================] - 0s 398us/step - loss: 0.4660 - accuracy: 0.8429
Epoch 71/200
3/3 [==============================] - 0s 368us/step - loss: 0.3960 - accuracy: 0.9393
Epoch 72/200
3/3 [==============================] - 0s 355us/step - loss: 0.5523 - accuracy: 0.6071
Epoch 73/200
3/3 [==============================] - 0s 361us/step - loss: 0.5266 - accuracy: 0.7821
Epoch 74/200
3/3 [==============================] - 0s 371us/step - loss: 0.4245 - accuracy: 0.9643
Epoch 75/200
3/3 [==============================] - 0s 367us/step - loss: 0.5024 - accuracy: 0.7786
Epoch 76/200
3/3 [==============================] - 0s 453us/step - loss: 0.3419 - accuracy: 0.9393
Epoch 77/200
3/3 [==============================] - 0s 405us/step - loss: 0.4930 - accuracy: 0.7429
Epoch 78/200
3/3 [==============================] - 0s 672us/step - loss: 0.3443 - accuracy: 0.9036
Epoch 79/200
3/3 [==============================] - 0s 386us/step - loss: 0.3864 - accuracy: 0.8893
Epoch 80/200
3/3 [==============================] - 0s 386us/step - loss: 0.3863 - accuracy: 0.9286
Epoch 81/200
3/3 [==============================] - 0s 391us/step - loss: 0.2771 - accuracy: 0.8679
Epoch 82/200
3/3 [==============================] - 0s 370us/step - loss: 0.6083 - accuracy: 0.5571
Epoch 83/200
3/3 [==============================] - 0s 387us/step - loss: 0.2801 - accuracy: 0.9393
Epoch 84/200
3/3 [==============================] - 0s 357us/step - loss: 0.2483 - accuracy: 0.9286
Epoch 85/200
3/3 [==============================] - 0s 355us/step - loss: 0.2511 - accuracy: 0.9643
Epoch 86/200
3/3 [==============================] - 0s 339us/step - loss: 0.3410 - accuracy: 0.8893
Epoch 87/200
3/3 [==============================] - 0s 361us/step - loss: 0.3432 - accuracy: 0.9036
Epoch 88/200
3/3 [==============================] - 0s 347us/step - loss: 0.3819 - accuracy: 0.8893
Epoch 89/200
3/3 [==============================] - 0s 361us/step - loss: 0.5142 - accuracy: 0.7179
Epoch 90/200
3/3 [==============================] - 0s 502us/step - loss: 0.3055 - accuracy: 0.9393
Epoch 91/200
3/3 [==============================] - 0s 377us/step - loss: 0.3144 - accuracy: 0.8536
Epoch 92/200
3/3 [==============================] - 0s 376us/step - loss: 0.3712 - accuracy: 0.9036
Epoch 93/200
3/3 [==============================] - 0s 389us/step - loss: 0.1974 - accuracy: 0.9393
Epoch 94/200
3/3 [==============================] - 0s 365us/step - loss: 0.3128 - accuracy: 0.9393
Epoch 95/200
3/3 [==============================] - 0s 376us/step - loss: 0.2194 - accuracy: 1.0000
Epoch 96/200
3/3 [==============================] - 0s 377us/step - loss: 0.1994 - accuracy: 1.0000
Epoch 97/200
3/3 [==============================] - 0s 360us/step - loss: 0.1734 - accuracy: 0.9643
Epoch 98/200
3/3 [==============================] - 0s 367us/step - loss: 0.1786 - accuracy: 1.0000
Epoch 99/200
3/3 [==============================] - 0s 358us/step - loss: 0.4158 - accuracy: 0.8286
Epoch 100/200
3/3 [==============================] - 0s 354us/step - loss: 0.3131 - accuracy: 0.7571
Epoch 101/200
3/3 [==============================] - 0s 350us/step - loss: 0.1953 - accuracy: 0.8893
Epoch 102/200
3/3 [==============================] - 0s 403us/step - loss: 0.2577 - accuracy: 0.8429
Epoch 103/200
3/3 [==============================] - 0s 417us/step - loss: 0.2648 - accuracy: 0.8893
Epoch 104/200
3/3 [==============================] - 0s 377us/step - loss: 0.2901 - accuracy: 0.8286
Epoch 105/200
3/3 [==============================] - 0s 383us/step - loss: 0.2822 - accuracy: 0.9393
Epoch 106/200
3/3 [==============================] - 0s 381us/step - loss: 0.2837 - accuracy: 0.9036
Epoch 107/200
3/3 [==============================] - 0s 382us/step - loss: 0.3064 - accuracy: 0.8536
Epoch 108/200
3/3 [==============================] - 0s 352us/step - loss: 0.3376 - accuracy: 0.9036
Epoch 109/200
3/3 [==============================] - 0s 376us/step - loss: 0.3412 - accuracy: 0.8536
Epoch 110/200
3/3 [==============================] - 0s 363us/step - loss: 0.1718 - accuracy: 1.0000
Epoch 111/200
3/3 [==============================] - 0s 347us/step - loss: 0.1899 - accuracy: 0.8786
Epoch 112/200
3/3 [==============================] - 0s 363us/step - loss: 0.2352 - accuracy: 0.8286
Epoch 113/200
3/3 [==============================] - 0s 373us/step - loss: 0.1378 - accuracy: 1.0000
Epoch 114/200
3/3 [==============================] - 0s 353us/step - loss: 0.4288 - accuracy: 0.7071
Epoch 115/200
3/3 [==============================] - 0s 456us/step - loss: 0.4202 - accuracy: 0.6821
Epoch 116/200
3/3 [==============================] - 0s 382us/step - loss: 0.2962 - accuracy: 0.8893
Epoch 117/200
3/3 [==============================] - 0s 394us/step - loss: 0.2571 - accuracy: 0.8893
Epoch 118/200
3/3 [==============================] - 0s 365us/step - loss: 0.2697 - accuracy: 1.0000
Epoch 119/200
3/3 [==============================] - 0s 358us/step - loss: 0.3102 - accuracy: 0.9036
Epoch 120/200
3/3 [==============================] - 0s 367us/step - loss: 0.2928 - accuracy: 0.8286
Epoch 121/200
3/3 [==============================] - 0s 374us/step - loss: 0.3157 - accuracy: 0.8286
Epoch 122/200
3/3 [==============================] - 0s 381us/step - loss: 0.3920 - accuracy: 0.7786
Epoch 123/200
3/3 [==============================] - 0s 335us/step - loss: 0.2090 - accuracy: 0.9036
Epoch 124/200
3/3 [==============================] - 0s 368us/step - loss: 0.5079 - accuracy: 0.7786
Epoch 125/200
3/3 [==============================] - 0s 337us/step - loss: 0.1900 - accuracy: 0.9393
Epoch 126/200
3/3 [==============================] - 0s 339us/step - loss: 0.2047 - accuracy: 0.9643
Epoch 127/200
3/3 [==============================] - 0s 479us/step - loss: 0.3705 - accuracy: 0.7679
Epoch 128/200
3/3 [==============================] - 0s 390us/step - loss: 0.1850 - accuracy: 0.9036
Epoch 129/200
3/3 [==============================] - 0s 642us/step - loss: 0.1594 - accuracy: 0.9393
Epoch 130/200
3/3 [==============================] - 0s 373us/step - loss: 0.2010 - accuracy: 0.8893
Epoch 131/200
3/3 [==============================] - 0s 369us/step - loss: 0.0849 - accuracy: 1.0000
Epoch 132/200
3/3 [==============================] - 0s 349us/step - loss: 0.1145 - accuracy: 1.0000
Epoch 133/200
3/3 [==============================] - 0s 360us/step - loss: 0.1796 - accuracy: 1.0000
Epoch 134/200
3/3 [==============================] - 0s 371us/step - loss: 0.2363 - accuracy: 0.8536
Epoch 135/200
3/3 [==============================] - 0s 386us/step - loss: 0.1922 - accuracy: 0.9393
Epoch 136/200
3/3 [==============================] - 0s 369us/step - loss: 0.3595 - accuracy: 0.7679
Epoch 137/200
3/3 [==============================] - 0s 369us/step - loss: 0.1506 - accuracy: 0.8893
Epoch 138/200
3/3 [==============================] - 0s 377us/step - loss: 0.2471 - accuracy: 0.8536
Epoch 139/200
3/3 [==============================] - 0s 417us/step - loss: 0.1768 - accuracy: 0.8536
Epoch 140/200
3/3 [==============================] - 0s 400us/step - loss: 0.2112 - accuracy: 0.9393
Epoch 141/200
3/3 [==============================] - 0s 377us/step - loss: 0.3652 - accuracy: 0.7179
Epoch 142/200
3/3 [==============================] - 0s 364us/step - loss: 0.3007 - accuracy: 0.8429
Epoch 143/200
3/3 [==============================] - 0s 361us/step - loss: 0.0518 - accuracy: 1.0000
Epoch 144/200
3/3 [==============================] - 0s 373us/step - loss: 0.2144 - accuracy: 0.8286
Epoch 145/200
3/3 [==============================] - 0s 353us/step - loss: 0.0888 - accuracy: 1.0000
Epoch 146/200
3/3 [==============================] - 0s 361us/step - loss: 0.1267 - accuracy: 1.0000
Epoch 147/200
3/3 [==============================] - 0s 341us/step - loss: 0.0321 - accuracy: 1.0000
Epoch 148/200
3/3 [==============================] - 0s 358us/step - loss: 0.0860 - accuracy: 1.0000
Epoch 149/200
3/3 [==============================] - 0s 375us/step - loss: 0.2151 - accuracy: 0.8893
Epoch 150/200
3/3 [==============================] - 0s 351us/step - loss: 0.1592 - accuracy: 1.0000
Epoch 151/200
3/3 [==============================] - 0s 531us/step - loss: 0.1450 - accuracy: 0.8786
Epoch 152/200
3/3 [==============================] - 0s 392us/step - loss: 0.1813 - accuracy: 0.9036
Epoch 153/200
3/3 [==============================] - 0s 404us/step - loss: 0.1197 - accuracy: 1.0000
Epoch 154/200
3/3 [==============================] - 0s 367us/step - loss: 0.0930 - accuracy: 1.0000
Epoch 155/200
3/3 [==============================] - 0s 580us/step - loss: 0.2587 - accuracy: 0.8893
Epoch 156/200
3/3 [==============================] - 0s 383us/step - loss: 0.0742 - accuracy: 1.0000
Epoch 157/200
3/3 [==============================] - 0s 353us/step - loss: 0.1197 - accuracy: 0.9643
Epoch 158/200
3/3 [==============================] - 0s 371us/step - loss: 0.1716 - accuracy: 0.8536
Epoch 159/200
3/3 [==============================] - 0s 337us/step - loss: 0.1300 - accuracy: 0.9643
Epoch 160/200
3/3 [==============================] - 0s 347us/step - loss: 0.1439 - accuracy: 0.9393
Epoch 161/200
3/3 [==============================] - 0s 366us/step - loss: 0.2597 - accuracy: 0.9393
Epoch 162/200
3/3 [==============================] - 0s 345us/step - loss: 0.1605 - accuracy: 0.8893
Epoch 163/200
3/3 [==============================] - 0s 468us/step - loss: 0.0437 - accuracy: 1.0000
Epoch 164/200
3/3 [==============================] - 0s 372us/step - loss: 0.0376 - accuracy: 1.0000
Epoch 165/200
3/3 [==============================] - 0s 391us/step - loss: 0.0474 - accuracy: 1.0000
Epoch 166/200
3/3 [==============================] - 0s 378us/step - loss: 0.3225 - accuracy: 0.7786
Epoch 167/200
3/3 [==============================] - 0s 368us/step - loss: 0.0770 - accuracy: 1.0000
Epoch 168/200
3/3 [==============================] - 0s 367us/step - loss: 0.5629 - accuracy: 0.7786
Epoch 169/200
3/3 [==============================] - 0s 359us/step - loss: 0.0177 - accuracy: 1.0000
Epoch 170/200
3/3 [==============================] - 0s 370us/step - loss: 0.1167 - accuracy: 1.0000
Epoch 171/200
3/3 [==============================] - 0s 349us/step - loss: 0.1313 - accuracy: 1.0000
Epoch 172/200
3/3 [==============================] - 0s 337us/step - loss: 0.0852 - accuracy: 0.9393
Epoch 173/200
3/3 [==============================] - 0s 375us/step - loss: 0.0545 - accuracy: 1.0000
Epoch 174/200
3/3 [==============================] - 0s 354us/step - loss: 0.0674 - accuracy: 0.9643
Epoch 175/200
3/3 [==============================] - 0s 355us/step - loss: 0.0911 - accuracy: 1.0000
Epoch 176/200
3/3 [==============================] - 0s 404us/step - loss: 0.0980 - accuracy: 0.9393
Epoch 177/200
3/3 [==============================] - 0s 396us/step - loss: 0.0465 - accuracy: 1.0000
Epoch 178/200
3/3 [==============================] - 0s 403us/step - loss: 0.1117 - accuracy: 0.9393
Epoch 179/200
3/3 [==============================] - 0s 373us/step - loss: 0.0415 - accuracy: 1.0000
Epoch 180/200
3/3 [==============================] - 0s 369us/step - loss: 0.0825 - accuracy: 1.0000
Epoch 181/200
3/3 [==============================] - 0s 425us/step - loss: 0.0378 - accuracy: 1.0000
Epoch 182/200
3/3 [==============================] - 0s 381us/step - loss: 0.1155 - accuracy: 0.9393
Epoch 183/200
3/3 [==============================] - 0s 354us/step - loss: 0.0207 - accuracy: 1.0000
Epoch 184/200
3/3 [==============================] - 0s 346us/step - loss: 0.0344 - accuracy: 1.0000
Epoch 185/200
3/3 [==============================] - 0s 379us/step - loss: 0.0984 - accuracy: 0.9393
Epoch 186/200
3/3 [==============================] - 0s 360us/step - loss: 0.1508 - accuracy: 0.8536
Epoch 187/200
3/3 [==============================] - 0s 361us/step - loss: 0.0463 - accuracy: 1.0000
Epoch 188/200
3/3 [==============================] - 0s 358us/step - loss: 0.0476 - accuracy: 0.9643
Epoch 189/200
3/3 [==============================] - 0s 379us/step - loss: 0.1592 - accuracy: 1.0000
Epoch 190/200
3/3 [==============================] - 0s 387us/step - loss: 0.0071 - accuracy: 1.0000
Epoch 191/200
3/3 [==============================] - 0s 405us/step - loss: 0.0527 - accuracy: 1.0000
Epoch 192/200
3/3 [==============================] - 0s 401us/step - loss: 0.0874 - accuracy: 0.9393
Epoch 193/200
3/3 [==============================] - 0s 355us/step - loss: 0.0199 - accuracy: 1.0000
Epoch 194/200
3/3 [==============================] - 0s 373us/step - loss: 0.1299 - accuracy: 0.9643
Epoch 195/200
3/3 [==============================] - 0s 360us/step - loss: 0.0929 - accuracy: 1.0000
Epoch 196/200
3/3 [==============================] - 0s 380us/step - loss: 0.0265 - accuracy: 1.0000
Epoch 197/200
3/3 [==============================] - 0s 358us/step - loss: 0.0843 - accuracy: 1.0000
Epoch 198/200
3/3 [==============================] - 0s 354us/step - loss: 0.0925 - accuracy: 1.0000
Epoch 199/200
3/3 [==============================] - 0s 327us/step - loss: 0.0770 - accuracy: 1.0000
Epoch 200/200
3/3 [==============================] - 0s 561us/step - loss: 0.0311 - accuracy: 1.0000
found in bag: 你好
[0 0 1 0 0 0 0 0 0 0 0 0 0 0]
found in bag: 请问
[('打招呼', '0.998965')]

瞬间执行完毕，秒杀intel芯片的mac，怎一个香字了得！

接下来，尝试安装另外一个在业界名声煊赫的深度学习框架Pytorch！

由于当前arm64架构的只支持3.9版本，所以我们来创建一个虚拟空间：

sudo conda create -n pytorch numpy matplotlib pandas python=3.9

这里提前将需要的基础库都一一安装，因为如果不在创建虚拟空间时提前安装，之后使用pip是安装不上的，安装成功后，激活环境：

(pytorch) ➜  conda activate pytorch                                        
(pytorch) ➜

随后下载arm64版本的pytorch安装包：https://github.com/wizyoung/AppleSiliconSelfBuilds/blob/main/builds/torch-1.8.0a0-cp39-cp39-macosx_11_0_arm64.whl

下载成功后，执行安装命令：

sudo pip install torch-1.8.0a0-cp39-cp39-macosx_11_0_arm64.whl

让我们来试试Pytorch在M1芯片加持后的性能，编写test_torch.py:

from tqdm import tqdm
import torch

@torch.jit.script
def foo():
    x = torch.ones((1024 * 12, 1024 * 12), dtype=torch.float32)
    y = torch.ones((1024 * 12, 1024 * 12), dtype=torch.float32)
    z = x + y
    return z


if __name__ == '__main__':
    z0 = None
    for _ in tqdm(range(10000000000)):
        zz = foo()
        if z0 is None:
            z0 = zz
        else:
            z0 += zz