Limit the default sequence length to 1024 for all models #1770
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
Pretrained models are supporting larger and larger sequence lengths. In gemma's case this is a particular nasty gotcha, as very few people have the compute resources to actually train on 8000 token long sequences. I think it might be the more user friendly approach to lower our default sequence length to 1024. This won't prohibit users from setting a longer sequence length, but it will lessen the unpleasant gotcha of suddenly using a ton of VRAM when training. We should still almost always document setting the sequence length in our code examples, as it's something the user generally should think about when fine-tuning or generating.
SamanehSaadat
approved these changes
Aug 14, 2024
|
Let's try it out. Something tells me we aren't done with discussions here, but hopefully this is a positive delta. |
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Outdated suggestions cannot be applied.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.
Pretrained models are supporting larger and larger sequence lengths. In gemma's case this is a particular nasty gotcha, as very few people have the compute resources to actually train on 8000 token long sequences.
I think it might be the more user friendly approach to lower our default sequence length to 1024. This won't prohibit users from setting a longer sequence length, but it will lessen the unpleasant gotcha of suddenly using a ton of VRAM when training.
We should still almost always document setting the sequence length in our code examples, as it's something the user generally should think about when fine-tuning or generating.