Add Seq Packing in NeMo / Neva2 #11633
Open
+611
−138
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
…a/neva2_seq_packing
Signed-off-by: yaoyu-33 <[email protected]>
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: yashaswikarnati <[email protected]>
…a/neva2_seq_packing # Conflicts: # nemo/collections/vlm/neva/model/base.py
Signed-off-by: yaoyu-33 <[email protected]>
…o yuya/neva2_seq_packing
Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
…a/neva2_seq_packing
Signed-off-by: yaoyu-33 <[email protected]>
# Conflicts: # nemo/collections/multimodal/data/energon/base.py # nemo/collections/multimodal/data/energon/conversation.py
Signed-off-by: yaoyu-33 <[email protected]>
…a/neva2_seq_packing
|
|
||
| __restore_key__: Tuple[Union[str, int, tuple], ...] = () | ||
| position_ids: torch.Tensor = field(default_factory=lambda: torch.empty(0, dtype=torch.float)) | ||
| packed_seq_params: PackedSeqParams = field(default_factory=lambda: PackedSeqParams()) |
Check notice
Code scanning / CodeQL
Unnecessary lambda Note
| """Sample type for image text raw batch""" | ||
|
|
||
| position_ids: torch.Tensor = field(default_factory=lambda: torch.empty(0, dtype=torch.float)) | ||
| packed_seq_params: PackedSeqParams = field(default_factory=lambda: PackedSeqParams()) |
Check notice
Code scanning / CodeQL
Unnecessary lambda Note
Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
cuichenx
reviewed
Jan 8, 2025
|
|
||
| __restore_key__: Tuple[Union[str, int, tuple], ...] = () | ||
| position_ids: torch.Tensor = field(default_factory=lambda: torch.empty(0, dtype=torch.float)) | ||
| packed_seq_params: PackedSeqParams = field(default_factory=lambda: PackedSeqParams()) |
Comment on lines
69
to
70
| packed_sequence=False, | ||
| packing_seq_length=4096, |
There was a problem hiding this comment.
can you use consistent naming as fine_tuning.py? i.e.
use one variable packed_sequence_size, and packed_sequence_size=-1 indicates not using packed sequence
There was a problem hiding this comment.
i kept packed_sequence here for consistency with other vlm data modules where we don't use packed_sequence_size
There was a problem hiding this comment.
changed name from packing_seq_length to packed_sequence_size
| @@ -1711,7 +1711,10 @@ def masked_token_loss(tensor: Tensor, mask: Tensor): | |||
| """ | |||
| losses = tensor.float() | |||
| loss_mask = mask.view(-1).float() | |||
| loss = torch.sum(losses.view(-1) * loss_mask) / loss_mask.sum() # sequence level nll | |||
| num_valid_tokens = loss_mask.sum() | |||
| if num_valid_tokens < 0.5: # no valid tokens | |||
There was a problem hiding this comment.
can you explain a bit more when this is the case? is this only valid for neva?
There was a problem hiding this comment.
not only for neva, also for SFT. If the system and user prompt is very long, and predict answer only. After truncation from right, there might not be any answer/valid tokens.
There was a problem hiding this comment.
Usually we truncate the input/context and keep answer intact, so that wouldn't happen
There was a problem hiding this comment.
yep the logic is a bit different for vlm. We don't want to truncate from left.
Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
…a/neva2_seq_packing # Conflicts: # nemo/collections/multimodal/data/energon/task_encoder.py
Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: Yu Yao <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
Signed-off-by: yaoyu-33 <[email protected]>
|
beep boop 🤖: 🙏 The following files have warnings. In case you are familiar with these, please try helping us to improve the code base. Your code was analyzed with PyLint. The following annotations have been identified: Mitigation guide:
By applying these rules, we reduce the occurance of this message in future. Thank you for improving NeMo's documentation! |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
What does this PR do ?
Add a one line overview of what this PR aims to accomplish.
Collection: [Note which collection this PR will affect]
Changelog
Usage
# Add a code snippet demonstrating how to use thisGitHub Actions CI
The Jenkins CI system has been replaced by GitHub Actions self-hosted runners.
The GitHub Actions CI will run automatically when the "Run CICD" label is added to the PR.
To re-run CI remove and add the label again.
To run CI on an untrusted fork, a NeMo user with write access must first click "Approve and run".
Before your PR is "Ready for review"
Pre checks:
PR Type:
If you haven't finished some of the above items you can still open "Draft" PR.
Who can review?
Anyone in the community is free to review the PR once the checks have passed.
Contributor guidelines contains specific people who can review PRs to various areas.
Additional Information