mirror/cube-studio
2
0
Fork 0
mirror of https://github.com/tencentmusic/cube-studio.git synced 2024-11-27 05:33:10 +08:00
Code Issues Packages Projects Releases Wiki Activity

add tensorboard

Browse Source
This commit is contained in:
cdllp2 2022-10-11 18:24:45 +08:00
parent d2e9c51049
commit fd47961487
5 changed files with 272 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
images/jupyter-notebook/deeplearning/Dockerfile
View File

@ -15,6 +15,11 @@ RUN pip install numpy pandas StatsModels Matplotlib Seaborn sklearn wheel SciPy
RUN pip install tensorflow keras torch torchvision keras nltk spacy gensim \
&& rm -rf /tmp/* /var/tmp/* /root/.cache
# 安装插件
RUN jupyter labextension install @jupyter-widgets/jupyterlab-manager
RUN pip install --upgrade jupyterlab-git && jupyter lab build && \
pip uninstall -y tensorboard && pip install tensorboard==2.10 jupyterlab-tensorboard-pro && jupyter labextension install jupyterlab_tensorboard_pro ; jupyter serverextension enable jupyterlab_tensorboard_pro --sys-prefix
# 拷贝examples
COPY examples/* /examples/

55
images/jupyter-notebook/readme.md Normal file
View File

@ -0,0 +1,55 @@
# notebook重启问题
关于续期：因为对gpu的占用方式为独占方式，所以对于gpu notebook会定时清理，需要按时续期。
关于清理：可以通过删除config.py中的delete_notebook定时任务，关闭掉定时清理notebook
关于环境：重启后会自动执行/mnt/$USERNAME/init.sh脚本，所以可以将环境写入此脚本，重启后自动安装环境，否则就需要打包到镜像或者离线anaconda文件
# 构建notebook镜像
需要构建新镜像并在生产上替换，才能让用户使用新的notebook镜像。
## 方法1：Dockerfile构建
jupyter镜像的构建在：https://github.com/tencentmusic/cube-studio/tree/master/images/jupyter-notebook
vscode镜像的构建在：https://github.com/tencentmusic/cube-studio/tree/master/images/theia
现在默认使用的镜像为
```
# notebook使用的镜像
NOTEBOOK_IMAGES=[
['ccr.ccs.tencentyun.com/cube-studio/notebook:vscode-ubuntu-cpu-base', 'vscode（cpu）'],
['ccr.ccs.tencentyun.com/cube-studio/notebook:vscode-ubuntu-gpu-base', 'vscode（gpu）'],
['ccr.ccs.tencentyun.com/cube-studio/notebook:jupyter-ubuntu-cpu-base', 'jupyter（cpu）'],
['ccr.ccs.tencentyun.com/cube-studio/notebook:jupyter-ubuntu-gpu-base','jupyter（gpu）'],
['ccr.ccs.tencentyun.com/cube-studio/notebook:jupyter-ubuntu-bigdata', 'jupyter（bigdata）'],
['ccr.ccs.tencentyun.com/cube-studio/notebook:jupyter-ubuntu-machinelearning', 'jupyter（machinelearning）'],
['ccr.ccs.tencentyun.com/cube-studio/notebook:jupyter-ubuntu-deeplearning', 'jupyter（deeplearning）'],
]
```
## 方法2，直接commit容器
也可以直接run一个容器，然后安装插件后将容器commmit成镜像。
```
# 启动jupyter
docker run --name jupyter -p 3000:3000 -d ccr.ccs.tencentyun.com/cube-studio/notebook:jupyter-ubuntu-cpu-base jupyter lab --notebook-dir=/ --ip=0.0.0.0 --no-browser --allow-root --port=3000 --NotebookApp.token='' --NotebookApp.password='' --NotebookApp.allow_origin='*'
# 启动vscode
docker run --name vscode -p 3000:3000 -d ccr.ccs.tencentyun.com/cube-studio/notebook:vscode-ubuntu-cpu-base node /home/theia/src-gen/backend/main.js /home/project --hostname=0.0.0.0 --port=3000
```
然后访问 http://xx.xx.xx.xx:3000/ ， web界面操作，安装notebook插件，安装pip/apt环境等。环境完整后，再使用如下命令commit成镜像。
```
docker commit notebook ccr.ccs.tencentyun.com/cube-studio/notebook:jupyter-ubuntu-cpu-base-1
```
# 修改配置文件
config.py中 NOTEBOOK_IMAGES 变量为notebook可选镜像。更新此变量即可。
# 其他类型的notebook
所有可提供在线编辑功能的web服务都可以定义为notebook。开发代码已提供对外，需要满足几个条件，可配置url前缀，用来区分不同的notebook。

61
install/kubernetes/mindspore/example/ms_wide_deep_dataparallel.yaml Normal file
View File

@ -0,0 +1,61 @@
# wide&deep dataparallel mode with scheduler and workers
# network : https://gitee.com/mindspore/models/tree/master/official/recommend/wide_and_deep
apiVersion: mindspore.gitee.com/v1
kind: MSJob
metadata:
name: ms-widedeep-dataparallel
namespace: pipeline
spec:
runPolicy:
cleanPodPolicy: None
successPolicy: AllWorkers
msReplicaSpecs:
Scheduler:
replicas: 1
restartPolicy: Never
template:
spec:
volumes:
- name: script-data
hostPath:
path: /home/fzh/wide_and_deep/
containers:
- name: mindspore
image: swr.cn-south-1.myhuaweicloud.com/mindspore-ci/mindspore-gpu:1.7.0-20220327121541
imagePullPolicy: IfNotPresent
command:
- /bin/bash
- -c
- python -s /home/fzh/wide_and_deep/train_and_eval_distribute.py --device_target="GPU" --epochs=1 --data_path=/home/fzh/wide_and_deep/criteo_mindrecord --batch_size=16000
volumeMounts:
- mountPath: /home/fzh/wide_and_deep/
name: script-data
env:
- name: GLOG_v
value: "1"
Worker:
replicas: 4
restartPolicy: Never
template:
spec:
volumes:
- name: script-data
hostPath:
path: /home/fzh/wide_and_deep/
containers:
- name: mindspore
image: swr.cn-south-1.myhuaweicloud.com/mindspore-ci/mindspore-gpu:1.7.0-20220327121541
imagePullPolicy: IfNotPresent
command:
- /bin/bash
- -c
- python -s /home/fzh/wide_and_deep/train_and_eval_distribute.py --device_target="GPU" --epochs=1 --data_path=/home/fzh/wide_and_deep/criteo_mindrecord --batch_size=16000
volumeMounts:
- mountPath: /home/fzh/wide_and_deep/
name: script-data
env:
- name: GLOG_v
value: "1"
# resources:
# limits:
# nvidia.com/gpu: 1

77
install/kubernetes/mindspore/example/ms_wide_deep_ps_distribute.yaml Normal file
View File

@ -0,0 +1,77 @@
# wide&deep ps mode distribute training.
# network : https://gitee.com/mindspore/models/tree/master/official/recommend/wide_and_deep
apiVersion: mindspore.gitee.com/v1
kind: MSJob
metadata:
name: ms-widedeep-ps-distribute
spec:
runPolicy:
cleanPodPolicy: None
successPolicy: AllWorkers
msReplicaSpecs:
Scheduler:
replicas: 1
restartPolicy: Never
template:
spec:
volumes:
- name: script-data
hostPath:
path: /home/fzh/wide_and_deep/
containers:
- name: mindspore
image: swr.cn-south-1.myhuaweicloud.com/mindspore-ci/mindspore-gpu:1.7.0-20220327121541
imagePullPolicy: IfNotPresent
command:
- /bin/bash
- -c
- python -s /home/fzh/wide_and_deep/train_and_eval_parameter_server_distribute.py --device_target="GPU" --epochs=1 --data_path=/home/fzh/wide_and_deep/criteo_mindrecord --parameter_server=1 --vocab_cache_size=300000
volumeMounts:
- mountPath: /home/fzh/wide_and_deep/
name: script-data
env:
- name: GLOG_v
value: "1"
PS:
replicas: 2
restartPolicy: Never
template:
spec:
volumes:
- name: script-data
hostPath:
path: /home/fzh/wide_and_deep/
containers:
- name: mindspore
image: swr.cn-south-1.myhuaweicloud.com/mindspore-ci/mindspore-gpu:1.7.0-20220327121541
imagePullPolicy: IfNotPresent
command:
- /bin/bash
- -c
- python -s /home/fzh/wide_and_deep/train_and_eval_parameter_server_distribute.py --device_target="GPU" --epochs=1 --data_path=/home/fzh/wide_and_deep/criteo_mindrecord --parameter_server=1 --vocab_cache_size=300000
volumeMounts:
- mountPath: /home/fzh/wide_and_deep/
name: script-data
Worker:
replicas: 4
restartPolicy: Never
template:
spec:
volumes:
- name: script-data
hostPath:
path: /home/fzh/wide_and_deep/
containers:
- name: mindspore
image: swr.cn-south-1.myhuaweicloud.com/mindspore-ci/mindspore-gpu:1.7.0-20220327121541
imagePullPolicy: IfNotPresent
command:
- /bin/bash
- -c
- python -s /home/fzh/wide_and_deep/train_and_eval_parameter_server_distribute.py --device_target="GPU" --epochs=1 --data_path=/home/fzh/wide_and_deep/criteo_mindrecord --parameter_server=1 --vocab_cache_size=300000 --sparse=False
volumeMounts:
- mountPath: /home/fzh/wide_and_deep/
name: script-data
resources:
limits:
nvidia.com/gpu: 1

74
install/kubernetes/mindspore/example/ms_wide_deep_ps_standalone.yaml Normal file
View File

@ -0,0 +1,74 @@
# wide&deep for ps mode which only has one worker training.
# network : https://gitee.com/mindspore/models/tree/master/official/recommend/wide_and_deep
apiVersion: mindspore.gitee.com/v1
kind: MSJob
metadata:
name: ms-widedeep-ps-standalone
spec:
runPolicy:
cleanPodPolicy: None
successPolicy: AllWorkers
msReplicaSpecs:
Scheduler:
replicas: 1
restartPolicy: Never
template:
spec:
volumes:
- name: script-data
hostPath:
path: /home/fzh/wide_and_deep/
containers:
- name: mindspore
image: swr.cn-south-1.myhuaweicloud.com/mindspore-ci/mindspore-gpu:1.7.0-20220327121541
imagePullPolicy: IfNotPresent
command:
- /bin/bash
- -c
- python -s /home/fzh/wide_and_deep/train_and_eval_parameter_server_standalone.py --device_target="GPU" --epochs=1 --data_path=/home/fzh/wide_and_deep/criteo_mindrecord --parameter_server=1 --vocab_cache_size=200000
volumeMounts:
- mountPath: /home/fzh/wide_and_deep/
name: script-data
env:
- name: GLOG_v
value: "1"
PS:
replicas: 2
restartPolicy: Never
template:
spec:
volumes:
- name: script-data
hostPath:
path: /home/fzh/wide_and_deep/
containers:
- name: mindspore
image: swr.cn-south-1.myhuaweicloud.com/mindspore-ci/mindspore-gpu:1.7.0-20220327121541
imagePullPolicy: IfNotPresent
command:
- /bin/bash
- -c
- python -s /home/fzh/wide_and_deep/train_and_eval_parameter_server_standalone.py --device_target="GPU" --epochs=1 --data_path=/home/fzh/wide_and_deep/criteo_mindrecord --parameter_server=1 --vocab_cache_size=200000
volumeMounts:
- mountPath: /home/fzh/wide_and_deep/
name: script-data
Worker:
replicas: 1
restartPolicy: Never
template:
spec:
volumes:
- name: script-data
hostPath:
path: /home/fzh/wide_and_deep/
containers:
- name: mindspore
image: swr.cn-south-1.myhuaweicloud.com/mindspore-ci/mindspore-gpu:1.7.0-20220327121541
imagePullPolicy: IfNotPresent
command:
- /bin/bash
- -c
- python -s /home/fzh/wide_and_deep/train_and_eval_parameter_server_standalone.py --device_target="GPU" --epochs=1 --data_path=/home/fzh/wide_and_deep/criteo_mindrecord --parameter_server=1 --vocab_cache_size=200000
volumeMounts:
- mountPath: /home/fzh/wide_and_deep/
name: script-data