lightning/Distributed training.md at cfbf305c9c344bc9b5933a527eb288d05d0b2616

4.1 KiB

Raw Blame History

Lightning makes multi-gpu training and 16 bit training trivial.

Note:
None of the flags below require changing anything about your lightningModel definition.

Choosing a backend

Lightning supports two backends. DataParallel and DistributedDataParallel. Both can be used for single-node multi-GPU training. For multi-node training you must use DistributedDataParallel.

You can toggle between each mode by setting this flag.

# DEFAULT uses DataParallel
trainer = Trainer(distributed_backend='dp')

# change to distributed data parallel
trainer = Trainer(distributed_backend='ddp')

If you request multiple nodes, the back-end will auto-switch to ddp. We recommend you use DistributedDataparallel even for single-node multi-GPU training. It is MUCH faster than DP but may have configuration issues depending on your cluster.

For a deeper understanding of what lightning is doing, feel free to read this guide.

16-bit mixed precision

16 bit precision can cut your memory footprint by half. If using volta architecture GPUs it can give a dramatic training speed-up as well.
First, install apex (if install fails, look here):

$ git clone https://github.com/NVIDIA/apex
$ cd apex
$ pip install -v --no-cache-dir --global-option="--cpp_ext" --global-option="--cuda_ext" ./

then set this use_amp to True.

# DEFAULT
trainer = Trainer(amp_level='O2', use_amp=False)

Single-gpu

Make sure you're on a GPU machine.

# set these flags
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
os.environ["CUDA_VISIBLE_DEVICES"] = "0"

# DEFAULT
trainer = Trainer(gpus=[0])

multi-gpu

Make sure you're on a GPU machine. You can set as many GPUs as you want. In this setting, the model will run on all 8 GPUs at once using DataParallel under the hood.

# set these flags
# lightning sets these flags for you automatically
# no need to set yourself
# os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
# os.environ["CUDA_VISIBLE_DEVICES"] = "0,1,2,3,4,5,6,7"


# to use DataParallel (default)
trainer = Trainer(gpus=[0,1,2,3,4,5,6,7], distributed_backend='dp')

# RECOMMENDED use DistributedDataParallel
trainer = Trainer(gpus=[0,1,2,3,4,5,6,7], distributed_backend='ddp')

Multi-node

Multi-node training is easily done by specifying these flags.

# train on 12*8 GPUs
trainer = Trainer(gpus=[0,1,2,3,4,5,6,7], nb_gpu_nodes=12)

In addition, make sure to set up your SLURM job correctly via the SlurmClusterObject. In particular, specify the number of tasks per node correctly.

cluster = SlurmCluster(
    hyperparam_optimizer=test_tube.HyperOptArgumentParser(),
    log_path='/some/path/to/save',
)

# OPTIONAL FLAGS WHICH MAY BE CLUSTER DEPENDENT
# which interface your nodes use for communication
cluster.add_command('export NCCL_SOCKET_IFNAME=^docker0,lo')

# see output of the NCCL connection process
# NCCL is how the nodes talk to each other
cluster.add_command('export NCCL_DEBUG=INFO')

# setting a master port here is a good idea.
cluster.add_command(f'export MASTER_PORT={PORT}')

# good to load the latest NCCL version
cluster.load_modules(['NCCL/2.4.7-1-cuda.10.0'])

# configure cluster
cluster.per_experiment_nb_nodes = 12 
cluster.per_experiment_nb_gpus = 8

cluster.add_slurm_cmd(cmd='ntasks-per-node', value=8, comment='1 task per gpu')

Finally, make sure to add a distributed sampler to your dataset. The distributed sampler copies a portion of your dataset onto each GPU. (World_size = gpus_per_node * nb_nodes).

# ie: this:
dataset = myDataset()
dataloader = Dataloader(dataset)

# becomes:
dataset = myDataset()
dist_sampler = torch.utils.data.distributed.DistributedSampler(dataset)
dataloader = Dataloader(dataset, sampler=dist_sampler)

Self-balancing architecture

Here lightning distributes parts of your module across available GPUs to optimize for speed and memory.

COMING SOON.

4.1 KiB Raw Blame History