Skip to content

GitLab

"vscode:/vscode.git/clone" did not exist on "05b0f1ea2f9dc012dbc19deabca7fa653db9a1ac"
Commit 20ff66a0 authored by Jeff Rasley's avatar Jeff Rasley Committed by GitHub
Browse files

Azure tutorial updates and cleanup (#43)

parent 4ee043cd
Show whitespace changes
Inline Side-by-side
Showing with 196 additions and 225 deletions
README.md
View file @ 20ff66a0
...@@ -10,10 +10,7 @@ efficient, and effective. ...@@ -10,10 +10,7 @@ efficient, and effective.
DeepSpeed can train DL models with over a hundred billion parameters on current DeepSpeed can train DL models with over a hundred billion parameters on current
generation of GPU clusters, while achieving over 5x in system performance generation of GPU clusters, while achieving over 5x in system performance
compared to the state-of-art. Early adopters of DeepSpeed have already produced compared to the state-of-art.
a language model (LM) with over 17B parameters establishing a new SOTA in the LM
category.
## Table of Contents ## Table of Contents
...@@ -87,10 +84,6 @@ replicated across data-parallel processes, ZeRO partitions model states to save ...@@ -87,10 +84,6 @@ replicated across data-parallel processes, ZeRO partitions model states to save
significant memory. The current implementation (stage 1 of ZeRO) reduces memory by up to significant memory. The current implementation (stage 1 of ZeRO) reduces memory by up to
4x relative to the state-of-art. You can read more about ZeRO in our [paper](https://arxiv.org/abs/1910.02054). 4x relative to the state-of-art. You can read more about ZeRO in our [paper](https://arxiv.org/abs/1910.02054).
With this impressive memory reduction, early adopters of DeepSpeed have already
produced alanguage model (LM) with over 17B parameters called
[Turing-NLG](link-to-turing-blog), establishing a new SOTA in the LM category.
### Scalability ### Scalability
DeepSpeed supports efficient data parallelism, model parallelism, and their DeepSpeed supports efficient data parallelism, model parallelism, and their
combination. ZeRO boosts the scaling capability and efficiency further. combination. ZeRO boosts the scaling capability and efficiency further.
...@@ -167,8 +160,9 @@ overview](./docs/features.md) for descriptions and usage. ...@@ -167,8 +160,9 @@ overview](./docs/features.md) for descriptions and usage.
### Installation ### Installation
Please see our [Azure tutorial](azure/README.md) to get started with DeepSpeed on Azure! * Please see our [Azure tutorial](docs/azure.md) to get started with DeepSpeed on Azure!
* If you're not on Azure we recommend using our docker image via `docker pull deepspeed/deepspeed:latest` which contains a pre-installed version of DeepSpeed and all the necessary dependencies.
* If you want to install DeepSpeed manually we provide an install script [install.sh](install.sh) to help install on a local machine or across an entire cluster.
### Writing DeepSpeed Models ### Writing DeepSpeed Models
DeepSpeed model training is accomplished using the DeepSpeed engine. The engine DeepSpeed model training is accomplished using the DeepSpeed engine. The engine
......
azure/Dockerfile deleted 100644 → 0
View file @ 4ee043cd
FROM nvidia/cuda:10.0-devel-ubuntu18.04
##############################################################################
# Installation/Basic Utilities
##############################################################################
RUN apt-get update && \
apt-get install -y --no-install-recommends \
software-properties-common \
openssh-client openssh-server \
pdsh curl sudo net-tools \
vim iputils-ping wget
##############################################################################
# Installation Latest Git
##############################################################################
RUN add-apt-repository ppa:git-core/ppa -y && \
apt-get update && \
apt-get install -y git && \
git --version
##############################################################################
# Python
##############################################################################
ENV DEBIAN_FRONTEND=noninteractive
ENV PYTHON_VERSION=3
RUN apt-get install -y python3 python3-dev && \
rm -f /usr/bin/python && \
ln -s /usr/bin/python3 /usr/bin/python && \
curl -O https://bootstrap.pypa.io/get-pip.py && \
python get-pip.py && \
rm get-pip.py && \
pip install --upgrade pip && \
# Print python an pip version
python -V && pip -V
##############################################################################
# TensorFlow
##############################################################################
ENV TENSORFLOW_VERSION=1.14.0
RUN pip install tensorflow-gpu==${TENSORFLOW_VERSION}
##############################################################################
# PyTorch
##############################################################################
ENV PYTORCH_VERSION=1.2.0
ENV TORCHVISION_VERSION=0.4.0
ENV TENSORBOARDX_VERSION=1.8
RUN pip install torch==${PYTORCH_VERSION}
RUN pip install torchvision==${TORCHVISION_VERSION}
RUN pip install tensorboardX==${TENSORBOARDX_VERSION}
##############################################################################
# Temporary Installation Directory
##############################################################################
ENV STAGE_DIR=/tmp
RUN mkdir -p ${STAGE_DIR}
##############################################################################
# Mellanox OFED
##############################################################################
ENV MLNX_OFED_VERSION=4.6-1.0.1.1
RUN apt-get install -y libnuma-dev
RUN cd ${STAGE_DIR} && \
wget -q -O - http://www.mellanox.com/downloads/ofed/MLNX_OFED-${MLNX_OFED_VERSION}/MLNX_OFED_LINUX-${MLNX_OFED_VERSION}-ubuntu18.04-x86_64.tgz | tar xzf - && \
cd MLNX_OFED_LINUX-${MLNX_OFED_VERSION}-ubuntu18.04-x86_64 && \
./mlnxofedinstall --user-space-only --without-fw-update --all -q && \
cd ${STAGE_DIR} && \
rm -rf ${STAGE_DIR}/MLNX_OFED_LINUX-${MLNX_OFED_VERSION}-ubuntu18.04-x86_64*
##############################################################################
# nv_peer_mem
##############################################################################
RUN mkdir -p ${STAGE_DIR} && \
git clone https://github.com/Mellanox/nv_peer_memory.git ${STAGE_DIR}/nv_peer_memory && \
cd ${STAGE_DIR}/nv_peer_memory && \
./build_module.sh && \
cd ${STAGE_DIR} && \
tar xzf ${STAGE_DIR}/nvidia-peer-memory_1.0.orig.tar.gz && \
cd ${STAGE_DIR}/nvidia-peer-memory-1.0 && \
apt-get install -y dkms && \
dpkg-buildpackage -us -uc && \
dpkg -i ${STAGE_DIR}/nvidia-peer-memory_1.0-8_all.deb
##############################################################################
## Ucomment and set SSH Daemon port
###############################################################################
ENV SSH_PORT=2222
RUN cat /etc/ssh/sshd_config > ${STAGE_DIR}/sshd_config && \
sed "0,/^#Port 22/s//Port ${SSH_PORT}/" ${STAGE_DIR}/sshd_config > /etc/ssh/sshd_config
##############################################################################
## Add deepspeed user
###############################################################################
# Add a deepspeed user with user id 8877
#RUN useradd --create-home --uid 8877 deepspeed
RUN useradd --create-home --uid 1000 --shell /bin/bash deepspeed
RUN usermod -aG sudo deepspeed
RUN echo "deepspeed ALL=(ALL) NOPASSWD: ALL" >> /etc/sudoers
# # Change to non-root privilege
USER deepspeed
##############################################################################
# DeepSpeed
# TODO: once repo is public we can install latest deepspeed via this command
##############################################################################
#RUN git clone https://github.com/microsoft/DeepSpeed.git ${STAGE_DIR}/DeepSpeed
#RUN cd ${STAGE_DIR}/DeepSpeed && \
# git checkout . && \
# git checkout master && \
# sudo ./install.sh
#RUN rm -rf ${STAGE_DIR}/DeepSpeed
#RUN python -c "import deepspeed; print(deepspeed.__version__)"
azure/README.md deleted 100644 → 0
View file @ 4ee043cd
# DeepSpeed with Azure
This tutorial will help you get started running DeepSpeed on [Azure VMs](https://azure.microsoft.com/en-us/services/virtual-machines/), support for [Azure ML](https://azure.microsoft.com/en-us/services/machine-learning/) will be coming soon!
To help with launching Azure instances we suggest using the [Azure CLI](https://docs.microsoft.com/en-us/cli/azure/?view=azure-cli-latest). We have created several helper scripts to get you quickly started using DeepSpeed with Azure.
* Install Azure CLI on your local box: https://docs.microsoft.com/en-us/cli/azure/install-azure-cli
* Alternatively you can use the Azure in-browser shell: https://shell.azure.com/
## Create an SSH key
Generate a SSH key that will be used across this tutorial to SSH into your VMs and between Docker containers. `ssh-keygen` is the recommended way of doing this. Our scripts assume your key is located inside the same directory as the Azure scripts.
## Azure Config JSON
Our helper scripts depend on the following a configuration JSON for deployment and setup. We have provided a simple example JSON (see: [azure_config.json](azure_config.json)) that sets up a basic environment with two VMs, see the example below.
```json
{
"num_vms": 2,
"location": "southcentralus",
"azure_sku": "Standard_NV6_Promo",
"ssh_private_key": "id_rsa",
"docker_ssh_port": 2222
}
```
## Create Azure VMs
[./create_vms.sh](create_vms.sh) will create VMs with the Azure SKU you choose and in the region you specify in your config JSON. Feel free to customize your JSON to your desired region/SKU. This step will take a few minutes to complete while it sets up all of your VMs on Azure.
## Setup VM environment to use DeepSpeed
[./setup_vms.sh](setup_vms.sh) will generate the MPI-style hostfile and SSH config that all of your VMs will use so that your Docker containers can talk to one another after they are setup.
## Start the DeepSpeed docker container
[./setup_docker.sh](setup_docker.sh) will pull the DeepSpeed docker image on all VMs and start a container instance in the background. This will take several minutes since it needs to pull the entire Docker image.
## Access VMs
[./azure_ssh.sh](azure_ssh.sh) will let you SSH into any of your VMs with this syntax: `./azure_ssh.sh <node-id> [command]`, the node-id is a number between [0, num_vms). This script will find the public IP address of your VM and use the SSH key you provided in the azure config JSON.
## Access DeepSpeed container
Everything should be up and running at this point, let's access the running DeepSpeed container on the first VM and make sure we can talk to the other containers in our setup. Let's complete the following steps:
* SSH into the first VM via: `./azure_ssh.sh 0`
* Change directories into the azure folder of this repo via: `cd ~/workdir/DeepSpeed/azure`
* Attach the running docker container via: `./attach.sh`
* You should now be able to ssh into any other docker container, the containers can be accessed via their SSH alias of 'worker-N' where N is the VM number between [0, num_vms). In this example we should be able to successfully run `ssh worker-1 hostname`.
## Run Cifar example model
TODO
azure/README.md 0 → 120000
View file @ 20ff66a0
../docs/azure.md
\ No newline at end of file
azure/attach.sh
View file @ 20ff66a0
#!/bin/bash #!/bin/bash
name=${1-deepspeed} name=${1-deepspeed}
docker exec -i -t $name /bin/bash docker exec -i -w /home/deepspeed -t $name /bin/bash
azure/build_docker_image.sh
View file @ 20ff66a0
#!/bin/bash #!/bin/bash
docker build -t deepspeed:0.1 -f Dockerfile . docker build -t deepspeed:0.1 -f ../Dockerfile .
azure/setup_docker.sh
View file @ 20ff66a0
...@@ -6,6 +6,13 @@ if [ ! -f ${azure_config} ]; then ...@@ -6,6 +6,13 @@ if [ ! -f ${azure_config} ]; then
exit 1 exit 1
fi fi
parallel=true
command -v pdsh
if [ $? != 0 ]; then
echo "Installing pdsh will allow for the docker pull to be done in parallel across the cluster. See: 'apt-get install pdsh'"
parallel=false
fi
ssh_key=`cat ${azure_config} | jq .ssh_private_key | sed 's/"//g'` ssh_key=`cat ${azure_config} | jq .ssh_private_key | sed 's/"//g'`
if [ $ssh_key == "null" ]; then echo 'missing ssh_private_key in config'; exit 1; fi if [ $ssh_key == "null" ]; then echo 'missing ssh_private_key in config'; exit 1; fi
num_vms=`cat ${azure_config} | jq .num_vms` num_vms=`cat ${azure_config} | jq .num_vms`
...@@ -13,9 +20,29 @@ if [ $num_vms == "null" ]; then echo 'missing num_vms in config'; exit 1; fi ...@@ -13,9 +20,29 @@ if [ $num_vms == "null" ]; then echo 'missing num_vms in config'; exit 1; fi
args="-i ${ssh_key} -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null" args="-i ${ssh_key} -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null"
username=deepspeed username=deepspeed
for node_id in `seq 0 $((num_vms - 1))`; do
update_script="
docker pull deepspeed/deepspeed:latest;
ln -s workdir/DeepSpeed/azure/attach.sh attach.sh;
cd workdir/DeepSpeed;
git pull;
git submodule update --init --recursive;
bash azure/start_container.sh;
"
if [ $parallel == true ]; then
echo "parallel docker pull"
hosts=""
for node_id in {0..1}; do
addr=`az vm list-ip-addresses | jq .[${node_id}].virtualMachine.network.publicIpAddresses[0].ipAddress | sed 's/"//g'`
hosts="${addr},${hosts}"
done
PDSH_SSH_ARGS_APPEND=${args} pdsh -w $hosts -l ${username} $update_script
else
echo "sequential docker pull"
for node_id in `seq 0 $((num_vms - 1))`; do
ip_addr=`az vm list-ip-addresses | jq .[${node_id}].virtualMachine.network.publicIpAddresses[0].ipAddress | sed 's/"//g'` ip_addr=`az vm list-ip-addresses | jq .[${node_id}].virtualMachine.network.publicIpAddresses[0].ipAddress | sed 's/"//g'`
addr=${username}@${ip_addr} addr=${username}@${ip_addr}
ssh ${args} $addr "docker pull deepspeed/deepspeed:latest" ssh ${args} $addr $update_script
ssh ${args} $addr "cd workdir/DeepSpeed; git pull; git submodule update --init --recursive; bash azure/start_container.sh" done
done fi
azure/setup_vms.sh
View file @ 20ff66a0
...@@ -33,14 +33,20 @@ for node_id in `seq 0 $((num_vms - 1))`; do ...@@ -33,14 +33,20 @@ for node_id in `seq 0 $((num_vms - 1))`; do
" >> ${ssh_config} " >> ${ssh_config}
done done
update_script="
sudo mkdir -p /job;
sudo chmod -R 777 /job;
mkdir -p workdir;
git clone https://github.com/microsoft/DeepSpeed.git workdir/DeepSpeed;
"
for node_id in `seq 0 $((num_vms - 1))`; do for node_id in `seq 0 $((num_vms - 1))`; do
ip_addr=`az vm list-ip-addresses | jq .[${node_id}].virtualMachine.network.publicIpAddresses[0].ipAddress | sed 's/"//g'` ip_addr=`az vm list-ip-addresses | jq .[${node_id}].virtualMachine.network.publicIpAddresses[0].ipAddress | sed 's/"//g'`
addr=${username}@${ip_addr} addr=${username}@${ip_addr}
echo "copying ssh keys, ssh config, hostfile to worker-${node_id}" echo "copying ssh keys, ssh config, hostfile to worker-${node_id}"
ssh $args ${addr} $update_script
scp $args ${ssh_key}* ${addr}:.ssh/ scp $args ${ssh_key}* ${addr}:.ssh/
scp $args ${ssh_config} ${addr}:.ssh/ scp $args ${ssh_config} ${addr}:.ssh/
ssh $args ${addr} "sudo mkdir -p /job/; sudo chmod -R 777 /job; mkdir -p workdir"
scp $args ${hostfile} ${addr}:/job/ scp $args ${hostfile} ${addr}:/job/
ssh $args ${addr} 'git clone https://github.com/microsoft/DeepSpeed.git workdir/DeepSpeed'
done done
rm $hostfile $ssh_config rm $hostfile $ssh_config
azure/shutdown_vms.sh
View file @ 20ff66a0
...@@ -16,8 +16,8 @@ resource_group=deepspeed_rg_$location ...@@ -16,8 +16,8 @@ resource_group=deepspeed_rg_$location
for i in `seq 0 $(( num_vms - 1))`; do for i in `seq 0 $(( num_vms - 1))`; do
vm_name=${base_vm_name}_$i vm_name=${base_vm_name}_$i
echo "shutting down $vm_name" echo "deallocating $vm_name"
az vm stop --resource-group $resource_group --name $vm_name az vm deallocate --resource-group $resource_group --name $vm_name
echo "deleting $vm_name" echo "deleting $vm_name"
az vm delete -y --resource-group $resource_group --name $vm_name az vm delete -y --resource-group $resource_group --name $vm_name
done done
bin/deepspeed.pt 0 → 120000
View file @ 20ff66a0
ds
\ No newline at end of file
bin/ds_ssh 0 → 100755
View file @ 20ff66a0
#!/bin/bash
# Copyright 2020 The Microsoft DeepSpeed Team
command -v pdsh
if [ $? != 0 ]; then
echo "Cannot find pdsh, please install via 'apt-get install -y pdsh'"
exit 1
fi
hostfile=/job/hostfile
if [ -f $hostfile ]; then
hosts=`cat $hostfile | awk '{print $1}' | paste -sd "," -`
export PDSH_RCMD_TYPE=ssh
pdsh -w ${hosts} $@
else
echo "Missing hostfile at ${hostfile}, executing command locally"
$@
fi
docs/azure.md 0 → 100644
View file @ 20ff66a0
# DeepSpeed with Azure
This tutorial will help you get started running DeepSpeed on [Azure virtual
machines](https://azure.microsoft.com/en-us/services/virtual-machines/). Support for
[Azure ML](https://azure.microsoft.com/en-us/services/machine-learning/) will be coming
soon!
To help with launching Azure instances we suggest using the [Azure
CLI](https://docs.microsoft.com/en-us/cli/azure/?view=azure-cli-latest). We have created
several helper scripts to get you quickly started using DeepSpeed with Azure.
* Install Azure CLI on your local box: https://docs.microsoft.com/en-us/cli/azure/install-azure-cli
* Alternatively you can use the Azure in-browser shell: https://shell.azure.com/
## Create an SSH key
Generate an SSH key that will be used across this tutorial to SSH into your VMs and
between Docker containers. `ssh-keygen` is the recommended way of doing this. Our scripts
assume your key is located inside the same directory as the Azure scripts.
## Azure Config JSON
Our helper scripts depend on the following a configuration JSON for deployment and setup.
We have provided a simple example JSON in `azure_config.json` that sets up a basic
environment with two VMs. See the example below:
```json
{
"num_vms": 2,
"location": "southcentralus",
"azure_sku": "Standard_NV6_Promo",
"ssh_private_key": "id_rsa",
"docker_ssh_port": 2222
}
```
## Dependencies
The scripts in this tutorial require [jq](https://stedolan.github.io/jq/) to help with
parsing JSON from the command line. Also it is recommended to install
[pdsh](https://linux.die.net/man/1/pdsh) to help launch ssh connections in parallel.
## Create Azure VMs
We first need to allocate the VMs. We provide a script
```bash
./create_vms.sh
```
to create VMs with the Azure SKU in the region specified in `azure_config.json`. Feel
free to customize your JSON to your desired region/SKU. This step will take a few minutes
to complete while it sets up all of your VMs on Azure.
## Setup VM environment to use DeepSpeed
Next, we need to configure the VM environment for DeepSpeed. We provide a script
```bash
./setup_vms.sh
```
to generate a [hostfile](../README.md#resource-configuration) and SSH
configuration on all of the VMs. This configuration will be used by the DeepSpeed
Docker containers in the next step.
## Start the DeepSpeed docker container
We now setup the DeepSpeed Docker containers on the VMs. We provide a script
```bash
./setup_docker.sh
```
to pull the DeepSpeed image onto all VMs and start a container instance in the
background. This will take several minutes since it needs to pull the entire Docker
image.
## Access VMs
The tool [azure_ssh.sh](azure_ssh.sh) will let you SSH into any of the VMs with this
syntax:
```bash
./azure_ssh.sh <node-id> [command]
```
where the `node-id` is a number between `0` and `num_vms-1`. This script will find the
public IP address of your VM and use the SSH key provided in the Azure configuration
JSON.
## Access DeepSpeed container
Everything should be up and running at this point. Set's access the running DeepSpeed
container on the first VM and make sure we can talk to the other containers in our setup.
Let's complete the following steps:
* SSH into the first VM via: `./azure_ssh.sh 0`
* Change directories into the azure folder of this repo via: `cd ~/workdir/DeepSpeed/azure`
* Attach the running docker container via: `./attach.sh`
* You should now be able to `ssh` into any other docker container, the containers can be
accessed via their SSH alias of `worker-N`, where `N` is the VM number between `0`
and `num_vms-1`. In this example we should be able to successfully run `ssh worker-1
hostname`. You can also use `ds_ssh` to execute a command in parallel on all of your
worker containers.
## Run CIFAR-10 example model
We will now run the DeepSpeed CIFAR-10 model example to test the VM setup. From inside
the first DeepSpeed container:
1) Install the python dependencies necessary to run the CIFAR-10 example model. You can
do this across your cluster via:
```bash
ds_ssh pip install -r ~/workdir/DeepSpeed/DeepSpeedExamples/cifar/requirements.txt
```
2) Now change directories to the CIFAR example:
```bash
cd ~/workdir/DeepSpeed/DeepSpeedExamples/cifar
```
3) Finally, launch training across all VMs:
```bash
deepspeed cifar10_deepspeed.py --deepspeed --deepspeed_config ds_config.json
```
Alternatively, we provide a helper script `./run_ds.sh`.
This will train a simple CIFAR-10 example model. The accuracy that you will achieve will
be dependent on the number of GPUs you are training with, we are using this example
simply to demonstrate that everything is setup correctly and less on training a suitable
CIFAR-10 model.
## Megatron-LM GPT2
DeepSpeed includes an example model using Megatron-LM's GPT2. Please refer to the full
[Megatron tutorial](tutorials/MegatronGPT2Tutorial.md) for more details.
* In order to fully train GPT2 with DeepSpeed and ZeRO we recommend using 8 instances of
Azure's Standard_ND40rs_v2 SKU for a total of 64 NVIDIA V100 GPUs. With this setup you
should be able to train 153.6 million samples in less than 2 weeks of training.
setup.py
View file @ 20ff66a0
...@@ -38,7 +38,9 @@ setup(name='deepspeed', ...@@ -38,7 +38,9 @@ setup(name='deepspeed',
"third_party", "third_party",
"csrc"]), "csrc"]),
scripts=['bin/deepspeed', scripts=['bin/deepspeed',
'bin/ds'], 'bin/deepspeed.pt',
'bin/ds',
'bin/ds_ssh'],
classifiers=['Programming Language :: Python :: 3.6'], classifiers=['Programming Language :: Python :: 3.6'],
ext_modules=ext_modules, ext_modules=ext_modules,
cmdclass=cmdclass) cmdclass=cmdclass)
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment