Activity
fixed missing import in test
fixed missing import in test
Force push
skip mqa cases
skip mqa cases
everything except mqa/gqa works
everything except mqa/gqa works
clean up a bit
clean up a bit
group size is a constexpr
group size is a constexpr
Force push
save
save
removed redundant file
removed redundant file
change USE_SPLIT to USE_SINGLE_BWD_KERNEL to make split default
change USE_SPLIT to USE_SINGLE_BWD_KERNEL to make split default
Force push
reenable
reenable
Minor fixes (#107)
Minor fixes (#107)
change USE_SPLIT to USE_SINGLE_BWD_KERNEL to make split default
change USE_SPLIT to USE_SINGLE_BWD_KERNEL to make split default
group size is a constexpr
group size is a constexpr
kinda working
kinda working
use do fp8 for dv
use do fp8 for dv
Force push
save
save
use do fp8 for dv
use do fp8 for dv
reduce diff
reduce diff
Force push
save
save
added envvar USE_SPLIT to toggle btw bwd kernels
added envvar USE_SPLIT to toggle btw bwd kernels
reduce diff
reduce diff
Deleted branch
Minor fixes (#107)
Minor fixes (#107)
Pull request merge
fixed dropout, esp w/ varlen
fixed dropout, esp w/ varlen
dv matches
dv matches
lse is good
lse is good
save clean up
save clean up
ref bug
ref bug
mark flakey test
mark flakey test
update readme
update readme
try again
try again