zzq96

hi, where can i find this paper's supplementary?

1

I have searched the AAAI digital library, but didnt find it . Can you provide the download link of this supplementary? thanks a lot!

in triton, tl.atomic_add use "rel_acq" memory semantics by default. after changing "rel_acq" to "relaxed", 2x faster of my flash attention triton code.

adding bias bwd makes fused attention bwd 30x slower, any suggestions?

1

I have written a bwd triton code , but I found that after adding the bias bwd , the speed is 30 times slower. The following is the time_weight bwd...

how to use atomic_add in shared memory?

3

I has a tensor, shape is [128,], we want to store this on shared memory and than use atomic_add to update this tensor, like this: ``` def test(output_ptr, input_ptr, input_index_ptr):...

generate card-style overview like Folder Note plugin

6

Folder Note plugin can show notes like this:  A card shows the title and a part of the notes This allows you to quickly recall the content of your...