Skip to content

[ROCm] Faster transformer inference on RDNA2 (gfx103x): dot2 attention, optional CK DPP GEMM - #1269

Open
zhaob1n wants to merge 4 commits into
lightvector:masterfrom
zhaob1n:rocm-rdna2-transformer
Open

zhaob1n wants to merge 4 commits into
lightvector:masterfrom
zhaob1n:rocm-rdna2-transformer

ROCm dot2 attention: fix NaN when a row's first key pair scores far b…

4fd62f5
Select commit
Loading
Failed to load commit list.