keras参考资料：https://keras.io/zh/getting-started/faq/#sample-batch-epoch

https://keras-cn.readthedocs.io/en/latest/

参考资料：https://www.jianshu.com/p/5d47f152ff62

https://blog.csdn.net/A632189007/article/details/77978058

###写法一：充分利用申请的资源

import tensorflow as tf

import keras.backend.tensorflow_backend as KTF

config = tf.ConfigProto()

config.gpu_options.allow_growth=True

session = tf.Session(config=config)

KTF.set_session(session)

#######写法二：限制GPU使用率

import os

import tensorflow as tf

import keras.backend.tensorflow_backend as KTF

config = tf.ConfigProto()

# 获取环境变量

env_dist = os.environ

#单卡GPU最大显存

card_gpu_mem_total = int(env_dist['ALIYUN_COM_GPU_MEM_CONTAINER'])

#启动Notebook分配的GPU显存

request_card_gpu_mem = int(env_dist['ALIYUN_COM_GPU_MEM_DEV'])

# percent=0.7 is because tensorflow control gpu memory is not accurate, it is recommended to multiply by 0.7 to ensure that the upper limit is not exceeded.

percent=0.7

fraction = round(card_gpu_mem_total*percent/request_card_gpu_mem, 1)

config.gpu_options.per_process_gpu_memory_fraction = fraction

session = tf.Session(config=config)

KTF.set_session(session)

##tensorflow调用GPU命名行

import os

import tensorflow as tf

# 获取环境变量

env_dist = os.environ

#单卡GPU最大显存

card_gpu_mem_total=int(env_dist['ALIYUN_COM_GPU_MEM_CONTAINER'])

#启动Notebook分配的GPU显存

request_card_gpu_mem=int(env_dist['ALIYUN_COM_GPU_MEM_DEV'])

# percent=0.7 is because tensorflow control gpu memory is not accurate, it is recommended to multiply by 0.7 to ensure that the upper limit is not exceeded.

percent=0.7

fraction=round(card_gpu_mem_total*percent/request_card_gpu_mem , 1)

config = tf.ConfigProto()

config.gpu_options.per_process_gpu_memory_fraction = fraction

sess = tf.Session(config=config)

# write your code

a = tf.constant(3)

b = tf.constant(4)

print('a + b = {0}'.format(sess.run(a+b)))

##pytorch调用GPU命令行

import os, torch

torch.cuda.is_available()

os.environ["CUDA_VISIBLE_DEVICES"] = os.getenv("ALIYUN_COM_GPU_MEM_IDX")

##pytorch按卡使用，不能按显存按需使用

###################

###keras调用GPU命令行

import os

os.environ["CUDA_VISIBLE_DEVICES"] = "0"

import tensorflow as tf

import keras.backend.tensorflow_backend as KTF

config = tf.ConfigProto()

config.gpu_options.allow_growth=True

session = tf.Session(config=config)

KTF.set_session(sess)

####查看GPU使用状态：nvidia-smi实时刷新并高亮显示状态

##其他参考资料：https://blog.csdn.net/huangfei711/article/details/79230446#

watch -n 1 -d nvidia-smi

####参考资料：https://blog.csdn.net/HaixWang/article/details/83349023

##https://blog.csdn.net/u013066730/article/details/77510015

##https://blog.csdn.net/silent56_th/article/details/60154637

##其他补充说明：

import os

import tensorflow as tf

import keras.backend.tensorflow_backend as KTF

#进行配置，每个GPU使用60%上限现存

os.environ["CUDA_VISIBLE_DEVICES"]="1,2" # 使用编号为1,2号的GPU

config = tf.ConfigProto()

config.gpu_options.per_process_gpu_memory_fraction = 0.6 #每个GPU现存上限控制在60%以内

session = tf.Session(config=config)

# 设置session

KTF.set_session(session)

##########

##分布式

##keras的分布式是利用TensorFlow实现的，要想完成分布式的训练，你需要将Keras注册在连接一个集群的TensorFlow会话上：

server = tf.train.Server.create_local_server()

sess = tf.Session(server.target)

from keras import backend as K

K.set_session(sess)

##########