Daniel Jünger
Daniel Jünger
I opened #683 and it seems that option 1 from my above post (regression after #672) is out of the race.
Thanks @kevkrist! Yeah, for this specific parameter combination this approach brings down the FPR a lot. However, it seems that it performs worse in other cases. > It might be...
/ok to test
/ok to test
@PointKernel the first overload was requested (or better proposed by me) in this thread (NV internal): https://nvidia.enterprise.slack.com/archives/CRJDZ0SKG/p1721975163922999 I think this overload is useful in general as it enables functionality similar...
/ok to test
The sketch is an array of `int32_t` with a pow2 size with only a few KB total. I think it's safe to expose it as a `span` or even `span`...
/ok to test 9c6722c843935d38362b3aa06dffdb1aed7ad811
/ok to test 9c6722c843935d38362b3aa06dffdb1aed7ad811
/ok to test 9c6722c843935d38362b3aa06dffdb1aed7ad811