Fix Qwen2.5-Omni get_chunked_index chunking functionality

[imkero]

What does this PR do?

This PR fixes the incorrect mrope position chunking function get_chunked_index in modular_qwen2_5_omni.

The shape of token_indices for modular_qwen2_5_omni.get_chunked_index is (3, seq_len) in main branch

It takes a constant value 3 (comes from len(token_indices)) in the loop condition to iterate the token_indices input (instead of the correct value token_indices.shape[1]). This will make it always produce only a single chunk. (Line 1160 vs Line 1166)

transformers/src/transformers/models/qwen2_5_omni/modular_qwen2_5_omni.py

Lines 1157 to 1168 in 27a25be

	def _iter():
	i, start_idx = 0, 0 # skip bos token
	current_chunk = 1
	while i < len(token_indices): # skip eos token
	if token_indices[0][i] - remove_index >= current_chunk * tokens_per_chunk:
	yield (start_idx, i)
	start_idx = i
	current_chunk += 1
	i += 1
	yield (start_idx, token_indices.shape[1])
	return list(_iter())

Another similar but correct impl lies in processing_qwen2_5_omni and we can take it as a reference. (Line 303 vs Line 309)

transformers/src/transformers/models/qwen2_5_omni/processing_qwen2_5_omni.py

Lines 300 to 311 in 27a25be

	def _iter():
	i, start_idx = 0, 0 # skip bos token
	current_chunk = 1
	while i < len(token_indices): # skip eos token
	if token_indices[i] >= current_chunk * tokens_per_chunk:
	yield (start_idx, i)
	start_idx = i
	current_chunk += 1
	i += 1
	yield (start_idx, len(token_indices))
	return list(_iter())

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

@BakerBunker can you take a look about this bugfix please?

[github-actions]

Hi 👋, thank you for opening this pull request! The pull request is converted to draft by default. The CI will be paused while the PR is in draft mode. When it is ready for review, please click the Ready for review button (at the bottom of the PR page). This will assign reviewers and trigger CI.

[BakerBunker]

Thank you, I will look at this tomorrow.

[BakerBunker]

LGTM, can be merged after the test cases are passed.

cc @zucchini-nlp

[imkero]

I think my change on the spatial_merge_size config broke some unit tests, will fix them soon.

[imkero]

I think my change on the spatial_merge_size config broke some unit tests, will fix them soon.

The failed tests have been fixed and pass now.

[BakerBunker]

@zucchini-nlp Ready to merge.

[zucchini-nlp]

Thanks, LGTM