Open source repositories tagged with #tensor-cores, ranked by health score.
🤖FFPA: Extends FlashAttention-2 via Split-D for large headdims, 1.5x~3×↑🎉 vs SDPA, up to 430T🎉 on H200.
