Skip to content
/ FfDL Public

Fabric for Deep Learning (FfDL, pronounced fiddle) is a Deep Learning Platform offering TensorFlow, Caffe, PyTorch etc. as a Service on Kubernetes

License

Notifications You must be signed in to change notification settings

IBM/FfDL

Folders and files

NameName
Last commit message
Last commit date

Latest commit

32aea41 · Dec 12, 2022

History

92 Commits
Apr 8, 2018
Sep 19, 2022
Aug 1, 2018
Aug 1, 2018
Apr 27, 2019
Apr 27, 2019
Jun 26, 2018
Dec 12, 2022
Apr 19, 2018
Sep 19, 2022
Apr 27, 2019
Aug 1, 2018
Oct 1, 2018
Aug 1, 2018
Jun 26, 2018
Aug 1, 2018
Jun 13, 2018
Aug 1, 2018
Aug 1, 2018
Aug 1, 2018
Sep 19, 2022
Jun 26, 2018
Sep 19, 2022
Jun 26, 2018
Sep 19, 2022
Apr 8, 2018
Jun 22, 2018
Sep 19, 2022
Feb 8, 2018
Aug 6, 2018
Jun 26, 2018
Aug 1, 2018
Jun 26, 2018

Repository files navigation

Read this in other languages: 中文.

build status

Fabric for Deep Learning (FfDL)

This repository contains the core services of the FfDL (Fabric for Deep Learning) platform. FfDL is an operating system "fabric" for Deep Learning. It is a collaboration platform for:

  • Framework-independent training of Deep Learning models on distributed hardware
  • Open Deep Learning APIs
  • Running Deep Learning hosting in user's private or public cloud

ffdl-architecture

To know more about the architectural details, please read the design document. If you are looking for demos, slides, collaterals, blogs, webinars and other materials related to FfDL, please find them here

Prerequisites

Usage Scenarios

  • If you are getting started and want to setup your own FfDL deployment, please follow the steps below.
  • If you have a FfDL deployment up and running, you can jump to FfDL User Guide to use FfDL for training your deep learning models.
  • If you want to leverage Jupyter notebooks to launch training on your FfDL cluster, please follow these instructions
  • If you have FfDL configured to use GPUs, and want to train using GPUs, follow steps here
  • To invoke Adversarial Robustness Toolbox to find vulnerabilities in your models, follow the instructions here
  • To deploy your trained models, follow the integration guide with Seldon
  • If you have used FfDL to train your models, and want to use a GPU enabled public cloud hosted service for further training and serving, please follow instructions here to train and serve your models using Watson Studio Deep Learning service.

Steps

  1. Quick Start
  1. Test
  2. Monitoring
  3. Development
  4. Clean Up
  5. Troubleshooting
  6. References

1. Quick Start

There are multiple installation paths for installing FfDL into an existing Kubernetes cluster. Below are the steps for quick install. If you want to follow more detailed step by step instructions , please visit the detailed installation guide

  • You need to initialize tiller with helm init before running the following commands.

1.1 Installation using Kubernetes Cluster

To install FfDL to any proper Kubernetes cluster, make sure kubectl points to the right namespace, then deploy the platform services:

export NAMESPACE=default # If your namespace does not exist yet, please create the namespace `kubectl create namespace $NAMESPACE` before running the make commands below
export SHARED_VOLUME_STORAGE_CLASS="ibmc-file-gold" # Change the storage class to what's available on your Cloud Kubernetes Cluster.

helm install ibmcloud-object-storage-plugin --name ibmcloud-object-storage-plugin --repo https://ibm.github.io/FfDL/helm-charts --set namespace=$NAMESPACE # Configure s3 driver on the cluster
helm install ffdl-helper --name ffdl-helper --repo https://ibm.github.io/FfDL/helm-charts --set namespace=$NAMESPACE,shared_volume_storage_class=$SHARED_VOLUME_STORAGE_CLASS --wait # Deploy all the helper micro-services for ffdl
helm install ffdl-core --name ffdl-core --repo https://ibm.github.io/FfDL/helm-charts --set namespace=$NAMESPACE,lcm.shared_volume_storage_class=$SHARED_VOLUME_STORAGE_CLASS --wait # Deploy all the core ffdl services.

1.2 Installation using Kubeadm-DIND

If you have Kubeadm-DIND installed on your machine, use these commands to deploy the FfDL platform:

export SHARED_VOLUME_STORAGE_CLASS=""
export NAMESPACE=default

./bin/s3_driver.sh # Copy the s3 drivers to each of the DIND node
helm install ibmcloud-object-storage-plugin --name ibmcloud-object-storage-plugin --repo https://ibm.github.io/FfDL/helm-charts --set namespace=$NAMESPACE,cloud=false
helm install ffdl-helper --name ffdl-helper --repo https://ibm.github.io/FfDL/helm-charts --set namespace=$NAMESPACE,shared_volume_storage_class=$SHARED_VOLUME_STORAGE_CLASS,localstorage=true --wait
helm install ffdl-core --name ffdl-core --repo https://ibm.github.io/FfDL/helm-charts --set namespace=$NAMESPACE,lcm.shared_volume_storage_class=$SHARED_VOLUME_STORAGE_CLASS --wait

# Forward the necessary microservices from the DIND cluster to your localhost.
./bin/dind-port-forward.sh

2. Test

To submit a simple example training job that is included in this repo (see etc/examples folder):

Note: For PUBLIC_IP, put down one of your Cluster Public IP that can access your Cluster's NodePorts. You can check your Cluster Public IP with kubectl get nodes -o wide. For IBM Cloud, you can get your Public IP with bx cs workers <cluster_name>.

export PUBLIC_IP=<Cluster Public IP> # Put down localhost if you are running with Kubeadm-DIND
make test-push-data-s3
make test-job-submit

3. Monitoring

The platform ships with a simple Grafana monitoring dashboard. The URL is printed out when running the status make target.

4. Development

Please refer to the developer guide for more details.

5. Clean Up

If you want to remove FfDL from your cluster, simply use the following commands.

helm delete --purge ffdl-core ffdl-helper

If you want to remove the storage driver from your cluster, run:

helm delete --purge ibmcloud-object-storage-plugin

For Kubeadm-DIND, you need to kill your forwarded ports. Note that the below command will kill all the ports that are created with kubectl.

kill $(lsof -i | grep kubectl | awk '{printf $2 " " }')

6. Troubleshooting

  • FfDL has only been tested under Mac OS and Linux
  • If glide install fails with an error complaining about non-existing paths (e.g., "Without src, cannot continue"), make sure to follow the standard Go directory layout (see Prerequisites section).

  • To remove FfDL on your Cluster, simply run make undeploy

  • When using the FfDL CLI to train a model, make sure your directory path doesn't have slashes / at the end.

  • If your job is stuck in pending stage, you can try to redeploy the plugin with helm install storage-plugin --set dind=true,cloud=false for Kubeadm-DIND and helm install storage-plugin for general Kubernetes Cluster. Also, double check your training job manifest file to make sure you have the correct object storage credentials.

7. References

Based on IBM Research work in Deep Learning.