Skip to content

Fixed RAdam path and showing progress during building features. #4

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 5 commits into
base: master
Choose a base branch
from
Open

Fixed RAdam path and showing progress during building features. #4

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
a1c82a1
Fixed RAdam import path.
prasad83 Jun 28, 2023
cfe5b2f
Show progress converting examples to features.
prasad83 Jun 28, 2023
94e4b42
Fixed torch.device non-negative index.
prasad83 Jun 28, 2023
4f8d751
Fixed pretrain modulo division by zero
prasad83 Jun 28, 2023
8e864fa
Fixed modulo division by zero in pretrain and evaluate
prasad83 Jun 28, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 3 additions & 2 deletions data_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,6 +5,7 @@
import torch
import torch.distributed as dist
from torch.utils.data import TensorDataset
from tqdm import tqdm

class PretrainInputExample:
"""A single example for unsupervised pre-training.
Expand Down Expand Up @@ -53,7 +54,7 @@ def convert_examples_to_features(examples,

# Create features
features = []
for i, example in enumerate(examples):
for i, example in enumerate(tqdm(examples)):
tokens = tokenizer.tokenize(example.text)
tokens = [bos_token] + tokens[:args.max_seq_len-2] + [eos_token] # BOS, EOS
tokens += [pad_token] * (args.max_seq_len - len(tokens))
Expand Down Expand Up @@ -116,4 +117,4 @@ def create_examples(args, tokenizer, mode='train'):
all_label_ids = torch.tensor([feature.label_id for feature in features], dtype=torch.long)
dataset = TensorDataset(all_input_ids, all_label_ids)

return dataset
return dataset
8 changes: 4 additions & 4 deletions trainer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,7 +6,7 @@
import torch.nn as nn
from torch.nn.parallel import DistributedDataParallel
from torch.utils.tensorboard import SummaryWriter
from radam import RAdam
import RAdam

from model import GPT, GPTLMHead, GPTClsHead

Expand Down Expand Up @@ -36,7 +36,7 @@ def __init__(self, args, train_loader, test_loader, tokenizer):
self.vocab_size = tokenizer.vocab_size
self.pad_id = tokenizer.pad_token_id
self.eos_id = tokenizer.eos_token_id
self.device = torch.device('cuda' if torch.cuda.is_available() and not args.no_cuda else 'cpu', args.local_rank)
self.device = torch.device('cuda' if torch.cuda.is_available() and not args.no_cuda else 'cpu', max(0, args.local_rank))
self.writer = SummaryWriter() if args.local_rank in [-1, 0] else None
self.n_gpus = torch.distributed.get_world_size() if args.distributed else torch.cuda.device_count()
assert args.pretrain != args.finetune # Do not set both finetune and pretrain arguments to the same (True, False)
Expand Down Expand Up @@ -101,7 +101,7 @@ def pretrain(self, epoch):

if self.args.local_rank in [-1, 0]:
self.writer.add_scalar('Loss/pre-train', loss.item(), ((epoch-1)*n_batches)+i)
if i % (n_batches//5) == 0 and i != 0:
if n_batches > 5 and i % (n_batches//5) == 0 and i != 0:
print('Iteration {} ({}/{})\tLoss: {:.4f}'.format(i, i, n_batches, losses/i))

print('Train Epoch {} [rank: {}]\t>\tLoss: {:.4f}'.format(epoch, self.args.local_rank, losses/n_batches))
Expand Down Expand Up @@ -134,7 +134,7 @@ def finetune(self, epoch):
if self.args.local_rank in [-1, 0]:
self.writer.add_scalar('Loss/fine-tune', loss.item(), ((epoch-1)*n_batches)+i)
self.writer.add_scalar('Accuracy/fine-tune', acc, ((epoch-1)*n_batches)+i)
if i % (n_batches//5) == 0 and i != 0:
if n_batches > 5 and i % (n_batches//5) == 0 and i != 0:
print('Iteration {} ({}/{})\tLoss: {:.4f} Acc: {:.1f}%'.format(i, i, n_batches, losses/i, accs/i*100.))

print('Train Epoch {} [rank: {}]\t>\tLoss: {:.4f} / Acc: {:.1f}%'.format(epoch, self.args.local_rank, losses/n_batches, accs/n_batches*100.))
Expand Down