[helion] backward support for swiglu #756
Merged
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
meta-cla
bot
added
the
CLA Signed
shunting314
force-pushed
the
swiglu-bwd
branch
from
8d8a27d to
99587ae
Compare
oulgen
reviewed
Oct 1, 2025
|
|
||
|
|
||
| @helion.kernel() | ||
| def swiglu_bwd(gout: Tensor, x1: Tensor, x2: Tensor) -> tuple[Tensor, Tensor]: |
There was a problem hiding this comment.
please add this to run.py there are two lists you need to update there
also please run with triton bench and generate perf/accuracy numbers
cc: @yf225
There was a problem hiding this comment.
@oulgen do you have an example to do that for a backward kernel? I can find a few examples for fwd but not bwd
There was a problem hiding this comment.
oh found 'rms_norm-bwd' in the run.py. will follow it
There was a problem hiding this comment.
I ran this command:
python benchmarks/run.py --metrics speedup,accuracy --kernel swiglu-bwd
but don't see the number for helion. Any ideas?
(B, T, H) liger_swiglu-speedup liger_swiglu-accuracy torch_compile_swiglu-speedup torch_compile_swiglu-accuracy
--------------- ---------------------- ----------------------- ------------------------------ -------------------------------
(4, 1024, 4096) 1.01139 1 1.03097 1
(4, 2048, 4096) 1.02854 1 1.00777 1
(4, 4096, 4096) 1.03631 1 1.03787 1
(4, 8192, 4096) 0.841614 1 1.04048 1
average 0.979463 1 1.02927 1
There was a problem hiding this comment.
I ran with the same command and with tritonbench's latest main (I did git pull in helion/benchmarks/tritonbench), and helion shows up:
(B, T, H) liger_swiglu-speedup liger_swiglu-accuracy torch_compile_swiglu-speedup torch_compile_swiglu-accuracy helion_swiglu_tritonbench-speedup helion_swiglu_tritonbench-accuracy
--------------- ---------------------- ----------------------- ------------------------------ ------------------------------- ----------------------------------- ------------------------------------
(4, 1024, 4096) 0.994532 1 0.992842 1 1.00311 1
(4, 2048, 4096) 0.950479 1 0.973353 1 0.844336 1
(4, 4096, 4096) 0.982585 1 1.02047 1 0.851285 1
(4, 8192, 4096) 1.01794 1 1.04066 1 0.977584 1
average 0.986385 1 1.00683 1 0.919078 1
@shunting314 it could be a tritonbench version issue - wonder would you like to try again? thanks!
There was a problem hiding this comment.
I can see the result now after pull from tritonbench
(B, T, H) liger_swiglu-speedup liger_swiglu-accuracy torch_compile_swiglu-speedup torch_compile_swiglu-accuracy helion_swiglu_tritonbench-speedup helion_swiglu_tritonbench-accuracy
--------------- ---------------------- ----------------------- ------------------------------ ------------------------------- ----------------------------------- ------------------------------------
(4, 1024, 4096) 1.06699 1 1.02534 1 1.00823 1
(4, 2048, 4096) 1.02478 1 0.952649 1 1.03361 1
(4, 4096, 4096) 0.991505 1 1.03377 1 1.02527 1
(4, 8192, 4096) 0.925007 1 1.06515 1 1.03323 1
average 1.00207 1 1.01923 1 1.02509 1
```
abbe582 to
5d29b48
Compare
74 tasks
oulgen
approved these changes
Oct 21, 2025
shunting314
force-pushed
the
swiglu-bwd
branch
from
5d29b48 to
9e19810
Compare
shunting314
force-pushed
the
swiglu-bwd
branch
from
9e19810 to
73d7d47
Compare
jansel
approved these changes
Oct 24, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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 the backward formula of swiglu in examples/swiglu.py