Skip to content

fix crashing #52

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
EYH0602 merged 3 commits into from
Feb 11, 2026
Merged

fix crashing #52

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
87edb9c
fix crashing
EYH0602 Feb 8, 2026
0bfd833
fix: disable safetensors saving to handle RoBERTa tied weights
EYH0602 Feb 9, 2026
57d54d5
Merge branch 'main' into fix/train-crashing
EYH0602 Feb 11, 2026
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
58 changes: 58 additions & 0 deletions modeling/dataloader.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -47,6 +47,35 @@ def contra_data_collator(mlm_collator, features):
code_batch = mlm_collator(code_features)
aug_batch = mlm_collator(aug_features)

# Pad to same seq_len (each batch is independently padded to its own max)
code_seq_len = code_batch["input_ids"].size(1)
aug_seq_len = aug_batch["input_ids"].size(1)
if code_seq_len != aug_seq_len:
pad_token_id = mlm_collator.tokenizer.pad_token_id
target_len = max(code_seq_len, aug_seq_len)
if code_seq_len < target_len:
pad = target_len - code_seq_len
code_batch["input_ids"] = torch.nn.functional.pad(
code_batch["input_ids"], (0, pad), value=pad_token_id
)
code_batch["attention_mask"] = torch.nn.functional.pad(
code_batch["attention_mask"], (0, pad), value=0
)
code_batch["labels"] = torch.nn.functional.pad(
code_batch["labels"], (0, pad), value=-100
)
else:
pad = target_len - aug_seq_len
aug_batch["input_ids"] = torch.nn.functional.pad(
aug_batch["input_ids"], (0, pad), value=pad_token_id
)
aug_batch["attention_mask"] = torch.nn.functional.pad(
aug_batch["attention_mask"], (0, pad), value=0
)
aug_batch["labels"] = torch.nn.functional.pad(
aug_batch["labels"], (0, pad), value=-100
)

# Combine batches
batch = {
"code_input_ids": code_batch["input_ids"],
Expand Down Expand Up @@ -138,6 +167,35 @@ def grouped_contra_data_collator(mlm_collator, max_num_augs, features):

aug_batch = mlm_collator(aug_features_flat)

# Pad to same seq_len (each batch is independently padded to its own max)
code_seq_len = code_batch["input_ids"].size(1)
aug_seq_len = aug_batch["input_ids"].size(1)
if code_seq_len != aug_seq_len:
pad_token_id = mlm_collator.tokenizer.pad_token_id
target_len = max(code_seq_len, aug_seq_len)
if code_seq_len < target_len:
pad = target_len - code_seq_len
code_batch["input_ids"] = torch.nn.functional.pad(
code_batch["input_ids"], (0, pad), value=pad_token_id
)
code_batch["attention_mask"] = torch.nn.functional.pad(
code_batch["attention_mask"], (0, pad), value=0
)
code_batch["labels"] = torch.nn.functional.pad(
code_batch["labels"], (0, pad), value=-100
)
else:
pad = target_len - aug_seq_len
aug_batch["input_ids"] = torch.nn.functional.pad(
aug_batch["input_ids"], (0, pad), value=pad_token_id
)
aug_batch["attention_mask"] = torch.nn.functional.pad(
aug_batch["attention_mask"], (0, pad), value=0
)
aug_batch["labels"] = torch.nn.functional.pad(
aug_batch["labels"], (0, pad), value=-100
)

batch = {
"code_input_ids": code_batch["input_ids"],
"code_attention_mask": code_batch["attention_mask"],
Expand Down
2 changes: 1 addition & 1 deletion modeling/pretrain.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -327,7 +327,7 @@ def main(
save_total_limit=3,
load_best_model_at_end=True,
dataloader_num_workers=max(1, (os.cpu_count() or 1) // _get_world_size()),
save_safetensors=False, # SplitHeadWrapper has tied weights from RobertaForMaskedLM
save_safetensors=False, # SplitHeadWrapper has tied weights from RoBERTa (embeddings ↔ lm_head); safetensors rejects shared tensors
)

trainer = ContrastiveTrainer(
Expand Down