# Tensorflow

[Kubeflow](https://github.com/google/kubeflow) 是 Google 发布的用于在 Kubernetes 集群中部署和管理 tensorflow 任务的框架。主要功能包括

* 用于管理 Jupyter 的 JupyterHub 服务
* 用于管理训练任务的 Tensorflow Training Controller
* 用于模型服务的 TF Serving 容器

## 部署

部署之前需要确保

* 一套部署好的 Kubernetes 集群或者 Minikube，并配置好 kubectl 命令行工具
* 安装 [ksonnet](https://github.com/ksonnet/ksonnet) 0.8.0 以上版本

对于开启 RBAC 的 Kubernetes 集群，首先要创建管理员角色绑定：

```
kubectl create clusterrolebinding tf-admin --clusterrole=cluster-admin --serviceaccount=default:tf-job-operator
```

然后运行以下命令部署

```bash
ks init my-kubeflow
cd my-kubeflow
ks registry add kubeflow github.com/google/kubeflow/tree/master/kubeflow
ks pkg install kubeflow/core
ks pkg install kubeflow/tf-serving
ks pkg install kubeflow/tf-job
ks generate core kubeflow-core --name=kubeflow-core
ks apply default -c kubeflow-core
```

如果有多个 Kubernetes 集群，也可以切换到其他其集群中部署，如

```bash
kubectl config use-context gke
ks env add gke
ks apply gke -c kubeflow-core
```

稍等一会，就可以看到 `tf-hub-lb` 服务的公网IP，也就是 JupyterHub 的访问地址

```bash
kubectl get svc tf-hub-lb
```

对于不支持 LoadBalancer Service 的集群，还可以通过端口转发（`http://127.0.0.1:8100`）的方式来访问：

```bash
kubectl port-forward tf-hub-0 8100:8000
```

JupyterHub 默认可以用任意用户名和密码登录。登陆后，可以使用自定义镜像来启动 Notebook Server，比如使用

* `gcr.io/kubeflow/tensorflow-notebook-cpu`
* `gcr.io/kubeflow/tensorflow-notebook-gpu`

## 训练示例

使用 CPU：

```bash
ks generate tf-cnn cnn --name=cnn
ks apply gke -c cnn
```

使用 GPU：

```bash
ks param set cnn num_gpus 1
ks param set  cnn num_workers 1
ks apply default -c cnn
```

## 模型部署

```
MODEL_COMPONENT=serveInception
MODEL_NAME=inception
MODEL_PATH=gs://cloud-ml-dev_jlewi/tmp/inception
ks generate tf-serving ${MODEL_COMPONENT} --name=${MODEL_NAME} --namespace=default --model_path=${MODEL_PATH}

ks apply gke -c ${MODEL_COMPONENT}
```

## 参考文档

* [Introducing Kubeflow - A Composable, Portable, Scalable ML Stack Built for Kubernetes](http://blog.kubernetes.io/2017/12/introducing-kubeflow-composable.html)
* <https://github.com/google/kubeflow>


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://kubernetes.feisky.xyz/practice/introduction/tensorflow.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.