Introduction || What is DDP || Single-Node Multi-GPU Training || Fault Tolerance || Multi-Node training || minGPT Training
Multinode Training¶
Authors: Suraj Subramanian
Launching multinode training jobs with
torchrunCode changes (and things to keep in mind) when moving from single-node to multinode training.
View the code used in this tutorial on GitHub
Familiarity with multi-GPU training and torchrun
2 or more TCP-reachable GPU machines (this tutorial uses AWS p3.2xlarge instances)
PyTorch installed with CUDA on all machines
Follow along with the video below or on youtube.
Multinode training involves deploying a training job across several machines. There are two ways to do this:
running a
torchruncommand on each machine with identical rendezvous arguments, ordeploying it on a compute cluster using a workload manager (like SLURM)
In this video we will go over the (minimal) code changes required to move from single-node multigpu to multinode training, and run our training script in both of the above ways.
Note that multinode training is bottlenecked by inter-node communication latencies. Running a training job on 4 GPUs on a single node will be faster than running it on 4 nodes with 1 GPU each.
Local and Global ranks¶
In single-node settings, we were tracking the
gpu_id of each device running our training process. torchrun tracks this value in an environment variable LOCAL_RANK
which uniquely identifies each GPU-process on a node. For a unique identifier across all the nodes, torchrun provides another variable
RANK which refers to the global rank of a process.
Warning
Do not use RANK for critical logic in your training job. When torchrun restarts processes after a failure or membership changes, there is no guarantee
that the processes will hold the same LOCAL_RANK and RANKS.
Heteregeneous Scaling¶
Torchrun supports heteregenous scaling i.e. each of your multinode machines can have different number of GPUs participating in the training job. In the video, I deployed the code on 2 machines where one machine has 4 GPUs and the other used only 2 GPUs.
Troubleshooting¶
Ensure that your nodes are able to communicate with each other over TCP.
Set env variable
NCCL_DEBUGtoINFO(usingexport NCCL_DEBUG=INFO) to print verbose logs that can help diagnose the issue.Sometimes you might need to explicitly set the network interface for the distributed backend (
export NCCL_SOCKET_IFNAME=eth0). Read more about this here.
Further Reading¶
Training a GPT model with DDP (next tutorial in this series)
Fault Tolerant distributed training (previous tutorial in this series)
