Skip to content

Fix group_by_modality_length in sync with LLaVA repo #90

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 1 commit into
base: main
Choose a base branch
from
Open

Fix group_by_modality_length in sync with LLaVA repo #90

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
6 changes: 3 additions & 3 deletions tinyllava/train/tinyllava_trainer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -43,12 +43,12 @@ def split_to_even_chunks(indices, lengths, num_chunks):
def get_modality_length_grouped_indices(lengths, batch_size, world_size, generator=None):
# We need to use torch for the random part as a distributed sampler will set the random seed for torch.
assert all(l != 0 for l in lengths), "Should not have zero length."
if all(l > 0 for l in lengths) or all(l < 0 for l in lengths):
# all samples are in the same modality
return get_length_grouped_indices(lengths, batch_size, world_size, generator=generator)
mm_indices, mm_lengths = zip(*[(i, l) for i, l in enumerate(lengths) if l > 0])
lang_indices, lang_lengths = zip(*[(i, -l) for i, l in enumerate(lengths) if l < 0])

assert len(mm_indices) > 0, "Should have at least one multimodal sample."
assert len(lang_indices) > 0, "Should have at least one language sample."

mm_shuffle = [mm_indices[i] for i in get_length_grouped_indices(mm_lengths, batch_size, world_size, generator=None)]
lang_shuffle = [lang_indices[i] for i in get_length_grouped_indices(lang_lengths, batch_size, world_size, generator=None)]
megabatch_size = world_size * batch_size
Expand Down