-
Notifications
You must be signed in to change notification settings - Fork 4
Update README and fix lint #23
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
@kwen2501 I've written a README and fixed some lint for releasing this at the PTC. Surya and I are working on the poster now (deadline is end of this week). Ping me if you want to talk about cleaning up this repo and what to put in the poster! |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM!
return torch_symm_mem_ag_mm | ||
if backend == "triton": | ||
return kraken.all_gather.all_gather_matmul | ||
return kraken.all_gather_fusion.all_gather_matmul |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think we can arrange ops in a couple categories:
kraken.comms
: pure comm ops, liketwo_shot_all_reduce
,all_to_all_on_device
, etc.kraken.fused
: here we put ops likeall_gather_matmul
kraken.quantized
: here we put ops with quantization, such as [mxfp8 moe training] mxfp8_on_device_all_to_all_v kernel pytorch/ao#3048
wdyt?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Agree!
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@danielvegamyhre we created a bucket for you -- kraken.quantized
:)
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
sweet! happy to add the kernel here as well, as a standalone building block/example.
any plans to import and use kraken in torchtitan?
…g soon) quantized kernels
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks a lot! Beautiful!
No description provided.