feat: flashv2 #94
Open
feat: flashv2 #94
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
|
Probably small performance increase is because we use custom attention masks |
egorchakov
requested changes
Apr 9, 2024
There was a problem hiding this comment.
we'd also probably wanna add a corresponding branch to https://github.com/yaak-ai/carGPT/blob/main/cargpt/components/llm.py#L229 (for attention viz at inference time)
cargpt/components/__init__.py
Outdated
| @@ -3,6 +3,8 @@ | |||
| from tensordict.tensorclass import _eq, _getitem # noqa: PLC2701 | |||
| from tensordict.utils import IndexType | |||
|
|
|||
| from cargpt.utils.attention import MemoryEfficientScaledDotProduct # noqa #type: ignore | |||
There was a problem hiding this comment.
do we actually need this exported here?
There was a problem hiding this comment.
We need to import it somewhere, so I find __init__.py the best choice
|
So let's just merge it without viz feature. Can still be used at full grown training after experimenting stage is completed (never) |
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Outdated suggestions cannot be applied.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.
Add flash attention v2
Basically, just extracted flash specific changes from https://github.com/yaak-ai/carGPT/tree/xformers-flashv2
Tested and it demonstrated running speed improvements: mean duration of train/val step (they are pretty similar) dropped by ~15% (1.9 -> 1.6 s)