Replace neural network encoder layer with fused LinearMax cuda kernel #112
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
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
daphne-cornelisse
changed the title
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
daphne-cornelisse
changed the title
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Outdated suggestions cannot be applied.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.
The neural network in pufferdrive is currently one of the main speed bottlenecks.
The linear + max operations are particularly expensive. This PR speeds up training by replacing the torch linear + max operations for the road and partner encoder with a cuda kernel that fuses these operations.
The result is a speed up of ~ 3.5 X in SPS. On an RTX-4080: 200K (main) -> 700K (new)
While the new network needs more steps, it is still an improvement over the net in main because identical performance is reached in less wall clock time.
I also switched the number of road points from 200 -> 128 and verified empirically that that is enough to get an off-road rate of near zero.