riaz.somc/transformers
1
0
Fork 0
mirror of https://github.com/huggingface/transformers.git synced 2025-07-31 02:02:21 +06:00
Code Issues Actions 33 Packages Projects Releases Wiki Activity

Optimize ForCausalLMLoss by removing unnecessary contiguous() call to reduce memory overhead (#35646)

Browse Source 
Optimize ForCausalLMLoss by removing unnecessary contiguous() calls to reduce memory overhead
This commit is contained in:
efsotr 2025-01-16 23:47:43 +08:00 committed by GitHub
parent 1302c32a84
commit 8ebe9d7166
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 4 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
src/transformers/loss/loss_utils.py
View File

@ -36,15 +36,15 @@ def ForCausalLMLoss(
logits = logits.float()
labels = labels.to(logits.device)
# Shift so that tokens < n predict n
shift_logits = logits[..., :-1, :].contiguous()
labels = nn.functional.pad(labels, (0, 1), value=ignore_index)
shift_labels = labels[..., 1:].contiguous()
# Flatten the tokens
shift_logits = shift_logits.view(-1, vocab_size)
logits = logits.view(-1, vocab_size)
shift_labels = shift_labels.view(-1)
# Enable model parallelism
shift_labels = shift_labels.to(shift_logits.device)
loss = fixed_cross_entropy(shift_logits, shift_labels, num_items_in_batch, ignore_index, **kwargs)
shift_labels = shift_labels.to(logits.device)
loss = fixed_cross_entropy(logits, shift_labels, num_items_in_batch, ignore_index, **kwargs)
return loss