[WIP] Add fast cuda kernels for one mode #154
Closed
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
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Outdated suggestions cannot be applied.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.
TODO: support seq mode
Add fast gemv kernel (based on https://github.com/wangsiping97/FastGEMV and added kernel fusion. ~10% faster than pytorch gemv) and a fused wkv_forward_one kernel (much faster, ~140 us -> 5us on 2080 1.5B model).
2080, 1.5B Model:
Main branch (blue bars represent CUDA kernels):
This branch:
FFN time: 359.77us -> 231.29us
ATT time: 291.71us -> 148.45us
Whole one mode time: 0.2487s -> 0.1409s
A100, 7B model:
Whole one mode time: 0.2657s -> 0.184s