LLaVaNeXT: pad on right if training #32134
Merged
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
amyeroberts
reviewed
Jul 22, 2024
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
|
Added a test @amyeroberts |
amyeroberts
reviewed
Jul 22, 2024
| @@ -43,6 +43,13 @@ The original code can be found [here](https://github.com/LLaVA-VL/LLaVA-NeXT/tre | |||
|
|
|||
| - We advise users to use `padding_side="left"` when computing batched generation as it leads to more accurate results. Simply make sure to call `processor.tokenizer.padding_side = "left"` before generating. | |||
|
|
|||
| <Tip warning={true}> | |||
|
|
|||
| - Llava-Next uses different number of patches for images and thus has to pad the inputs inside modeling code, aside from the padding done when processing the inputs. The default setting is "left-padding" if model is in `eval()` mode, otherwise "right-padding". | |||
amyeroberts
approved these changes
Jul 22, 2024
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Outdated suggestions cannot be applied.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?
Fixes #32112. The issue didn't mention any bug, as it's using batch-size=1 for training but it led me to this PR. This way we are more confident on which side to pad the inputs. I verified the padding is "right" when training with Trainer and in inference it runs to the block where we infer padding from attn mask