Changed the way that the memory for the activations is calculated#6
Open
krvas wants to merge 2 commits intomanuelescobar-dev:mainfrom
Open
Changed the way that the memory for the activations is calculated#6krvas wants to merge 2 commits intomanuelescobar-dev:mainfrom
krvas wants to merge 2 commits intomanuelescobar-dev:mainfrom
Conversation
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Hi,
I noticed that you had pulled a formula for calculating the amount of memory required from this paper. However, the paper makes certain assumptions about what precision floats we are using for the model, and on that basis it comes up with the formula. The code here is made to work with any kind of precision, so I made some edits that make the same formula from the paper more generalised. Hope this helps!
I also realised that KV cache memory is being calculated for training. However, KV caching is only done during inference, so I have removed that from the memory calculation.