def make_model_and_data(args, device, new_data: bool = True):
device = torch.device(
"cuda") if torch.cuda.is_available() else torch.device("cpu")
if new_data:
vocab_size = 10000
model, criterion, optimizer, scaler = make_model(
args, device, vocab_size)
lm_dataset = BenchmarkLMDataset()
lm_dataloader = DataLoader(lm_dataset,
batch_size=args.batch_size,
shuffle=True,
num_workers=0,
collate_fn=collate_sentences_lm)
return {
"model": model,
"criterion": criterion,
"optimizer": optimizer,
"data": lm_dataloader,
"vocab_size": vocab_size,
}
else:
data = get_data(device)
ntokens, train_data, val_data, test_data = data
model, criterion, optimizer, scaler = make_model(args, device, ntokens)
return {
"model": model,
"criterion": criterion,
"optimizer": optimizer,
"data": data,
}
def get_synthetic_dataloader(args):
"""Returns dataloader for synthetic data."""
if args.model_name == "lm":
lm_dataset = BenchmarkLMDataset()
lm_dataloader = DataLoader(
lm_dataset, batch_size=args.batch_size, shuffle=True, num_workers=0, collate_fn=collate_sentences_lm
)
return lm_dataloader
else:
raise RuntimeError("Unrecognized args.model_mame " % args.model_name)
def make_model_and_data(args, device):
device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
vocab_size = 10000
model, criterion, optimizer = make_model(args, device, vocab_size)
lm_dataset = BenchmarkLMDataset()
lm_dataloader = DataLoader(
lm_dataset, batch_size=args.batch_size, shuffle=True, num_workers=0, collate_fn=collate_sentences_lm
)
return {
"model": model,
"criterion": criterion,
"optimizer": optimizer,
"data": lm_dataloader,
"vocab_size": vocab_size,
}
