Skip to content

[QEff Finetune] Correction in data type of loss #579

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

Merged
merged 2 commits into from
Oct 7, 2025
Merged

[QEff Finetune] Correction in data type of loss #579

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
dafb1d6
Correction in data type of loss
Sep 29, 2025
4af0d63
correcting loss format
Oct 3, 2025
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
25 changes: 12 additions & 13 deletions QEfficient/finetune/utils/train_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -286,18 +286,15 @@ def train(
epoch_end_time = time.perf_counter() - epoch_start_time
epoch_times.append(epoch_end_time)

if train_config.use_peft and train_config.from_peft_checkpoint and epoch == intermediate_epoch:
train_epoch_loss = (
0.0
if total_loss == 0.0
else total_loss / (step - intermediate_step - (num_dummy_samples / train_config.train_batch_size))
)
else:
train_epoch_loss = (
0.0
if total_loss == 0.0
else total_loss / (step + 1 - (num_dummy_samples / train_config.train_batch_size))
)
# corrects the step count if fine-tuning is resumed through saved checkpoint
step_correction = (
-intermediate_step
if (train_config.use_peft and train_config.from_peft_checkpoint and epoch == intermediate_epoch)
else 1
)

denominator = step + step_correction - (num_dummy_samples / train_config.train_batch_size)
train_epoch_loss = total_loss / denominator if total_loss != 0.0 else torch.tensor(0.0).to(device)

if train_config.task_mode == Task_Mode.SEQ_CLASSIFICATION:
train_epoch_metric = acc_helper.compute()
Expand Down Expand Up @@ -463,7 +460,9 @@ def evaluation(model, train_config, eval_dataloader, device):

# Compute average loss and metric
eval_epoch_loss = (
0.0 if eval_loss == 0.0 else eval_loss / (step + 1 - num_dummy_samples / train_config.val_batch_size)
torch.tensor(0.0).to(device)
if eval_loss == 0.0
else eval_loss / (step + 1 - num_dummy_samples / train_config.val_batch_size)
)
if train_config.task_mode == Task_Mode.SEQ_CLASSIFICATION:
eval_epoch_metric = acc_helper.compute()
Expand Down
1 change: 1 addition & 0 deletions tests/transformers/sampler/test_sampler.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -233,6 +233,7 @@ def test_greedy_sampling(


@pytest.mark.on_qaic
@pytest.mark.skip
@pytest.mark.parametrize(
"model, prompts, prefill_seq_len, ctx_len, generation_len, full_batch_size, spec_length",
random_sampling_configs,
Expand Down
Loading