I am training on 3 servers using distributed data parallelism with 1 gpu on each server. I have 3 GPUs in total. How can I profile such a training? Can I collect and analyze each worker’s data such as running times, memory status on the master?
Here is my trainer script:
import torch
import torch.nn.functional as F
import os
import time
import psutil
import argparse
from torch.utils.data import DataLoader
from dataloader import MyDataset
from wideresnet import build_wideresnet
from torch.utils.data.distributed import DistributedSampler
from torch.nn.parallel import DistributedDataParallel as DDP
from torch.distributed import init_process_group, destroy_process_group
def ddp_setup():
init_process_group(backend="nccl")
class Trainer:
def __init__(
self,
model: torch.nn.Module,
train_data: DataLoader,
optimizer: torch.optim.Optimizer,
save_every: int,
snapshot_path: str
) -> None:
self.local_rank = int(os.environ["LOCAL_RANK"])
self.global_rank = int(os.environ["RANK"])
self.model = model.to(self.local_rank)
self.train_data = train_data
self.optimizer = optimizer
self.save_every = save_every
self.epochs_run = 0
self.snapshot_path = snapshot_path
if os.path.exists(snapshot_path):
print("Loading snapshot")
self._load_snapshot(snapshot_path)
self.model = DDP(self.model, device_ids=[self.local_rank], find_unused_parameters=True)
def _load_snapshot(self, snapshot_path):
loc = f"cuda:{self.gpu_id}"
snapshot = torch.load(snapshot_path, map_location=loc)
self.model.load_state_dict(snapshot["MODEL_STATE"])
self.epochs_run = snapshot["EPOCHS_RUN"]
print(f"Resuming training from snapshot at Epoch {self.epochs_run}")
def _save_snapshot(self, epoch):
snapshot = {
"MODEL_STATE": self.model.module.state_dict(),
"EPOCHS_RUN": epoch,
}
torch.save(snapshot, self.snapshot_path)
print(f"Epoch {epoch} | Training snapshot saved at {self.snapshot_path}")
def train(self, max_epochs: int):
for epoch in range(self.epochs_run, max_epochs):
b_sz = len(next(iter(self.train_data))[0])
print(f"[GPU{self.global_rank}] Epoch {epoch} | Batchsize: {b_sz} | Steps: {len(self.train_data)}")
self.train_data.sampler.set_epoch(epoch)
with torch.profiler.profile(activities=[torch.profiler.ProfilerActivity.CPU, torch.profiler.ProfilerActivity.CUDA], profile_memory=True,
on_trace_ready=torch.profiler.tensorboard_trace_handler(f"runs/gpu{self.global_rank}"), ) as prof:
for b_id ,(source, targets) in enumerate(self.train_data):
source = source.to(self.local_rank)
targets = targets.to(self.local_rank)
self.optimizer.zero_grad()
output = self.model(source)
loss = F.cross_entropy(output, targets)
loss.backward()
self.optimizer.step()
prof.step()
#if self.local_rank == 0 and epoch % self.save_every == 0:
# self._save_snapshot(epoch)
def load_train_objs(args):
train_set = MyDataset(data_dir=args.data_dir, annotations=args.annotations, num_classes=args.num_classes) # load your dataset
model = build_wideresnet(depth=28, widen_factor=4, dropout=0, num_classes=args.num_classes) # load your model
optimizer = torch.optim.SGD(model.parameters(), lr=1e-3)
return train_set, model, optimizer
def main(args, save_every: int, total_epochs: int, batch_size: int, snapshot_path: str = "snapshot.pt"):
ddp_setup()
dataset, model, optimizer = load_train_objs(args)
train_data = DataLoader(dataset, batch_size=batch_size, pin_memory=True, shuffle=False, sampler=DistributedSampler(dataset))
trainer = Trainer(model, train_data, optimizer, save_every, snapshot_path)
trainer.train(total_epochs)
destroy_process_group()
if __name__ == "__main__":
parser = argparse.ArgumentParser(description='distributed training job')
parser.add_argument('total_epochs', type=int, help='Total epochs to train the model')
parser.add_argument('save_every', type=int, help='How often to save a snapshot')
parser.add_argument('--batch_size', default=50, type=int, help='Input batch size on each device (default: 32)')
args = parser.parse_args()
args.data_dir = "/media/data-science"
args.annotations = "annotations.csv"
args.num_classes = 20
main(args,args.save_every, args.total_epochs, args.batch_size)