Tri Dao ea7a98f15d Fix backward with softcap il y a 2 mois
..
__init__.py 7f67966cc7 FA3 initial code release il y a 5 mois
benchmark_attn.py 3669b25206 bwd benchmark + small fixes (#1129) il y a 4 mois
benchmark_flash_attention_fp8.py c92ca63268 FA3 FP8 qkv descales + restore max offset for h128 causal + added sync for producer WG (#1173) il y a 3 mois
epilogue_bwd_sm90_tma.hpp 6e8b25e426 Refactor il y a 2 mois
epilogue_fwd_sm90_tma.hpp 6e8b25e426 Refactor il y a 2 mois
flash.h 6e8b25e426 Refactor il y a 2 mois
flash_api.cpp ea7a98f15d Fix backward with softcap il y a 2 mois
flash_attn_interface.py 6e8b25e426 Refactor il y a 2 mois
flash_bwd_hdim128_bf16_sm90.cu bafe253042 [FA3] Bwd il y a 4 mois
flash_bwd_hdim128_fp16_sm90.cu 7f67966cc7 FA3 initial code release il y a 5 mois
flash_bwd_hdim192_bf16_sm90.cu 6e8b25e426 Refactor il y a 2 mois
flash_bwd_hdim192_fp16_sm90.cu 6e8b25e426 Refactor il y a 2 mois
flash_bwd_hdim256_bf16_sm90.cu 6e8b25e426 Refactor il y a 2 mois
flash_bwd_hdim256_fp16_sm90.cu 7f67966cc7 FA3 initial code release il y a 5 mois
flash_bwd_hdim64_bf16_sm90.cu bafe253042 [FA3] Bwd il y a 4 mois
flash_bwd_hdim64_fp16_sm90.cu 7f67966cc7 FA3 initial code release il y a 5 mois
flash_bwd_hdim96_bf16_sm90.cu bafe253042 [FA3] Bwd il y a 4 mois
flash_bwd_hdim96_fp16_sm90.cu bafe253042 [FA3] Bwd il y a 4 mois
flash_bwd_kernel.h 6e8b25e426 Refactor il y a 2 mois
flash_bwd_launch_template.h ea7a98f15d Fix backward with softcap il y a 2 mois
flash_bwd_postprocess_kernel.h 6e8b25e426 Refactor il y a 2 mois
flash_bwd_preprocess_kernel.h 6e8b25e426 Refactor il y a 2 mois
flash_fwd_hdim128_bf16_sm90.cu 74b0761ff7 [FA3] BF16 forward il y a 5 mois
flash_fwd_hdim128_e4m3_sm90.cu 6e8b25e426 Refactor il y a 2 mois
flash_fwd_hdim128_fp16_sm90.cu 7f67966cc7 FA3 initial code release il y a 5 mois
flash_fwd_hdim192_bf16_sm90.cu 6e8b25e426 Refactor il y a 2 mois
flash_fwd_hdim192_e4m3_sm90.cu 6e8b25e426 Refactor il y a 2 mois
flash_fwd_hdim192_fp16_sm90.cu 6e8b25e426 Refactor il y a 2 mois
flash_fwd_hdim256_bf16_sm90.cu 74b0761ff7 [FA3] BF16 forward il y a 5 mois
flash_fwd_hdim256_e4m3_sm90.cu 6e8b25e426 Refactor il y a 2 mois
flash_fwd_hdim256_fp16_sm90.cu 7f67966cc7 FA3 initial code release il y a 5 mois
flash_fwd_hdim64_bf16_sm90.cu 74b0761ff7 [FA3] BF16 forward il y a 5 mois
flash_fwd_hdim64_e4m3_sm90.cu 6e8b25e426 Refactor il y a 2 mois
flash_fwd_hdim64_fp16_sm90.cu 7f67966cc7 FA3 initial code release il y a 5 mois
flash_fwd_hdim96_bf16_sm90.cu 6e8b25e426 Refactor il y a 2 mois
flash_fwd_hdim96_e4m3_sm90.cu 6e8b25e426 Refactor il y a 2 mois
flash_fwd_hdim96_fp16_sm90.cu 6e8b25e426 Refactor il y a 2 mois
flash_fwd_kernel.h 6e8b25e426 Refactor il y a 2 mois
flash_fwd_launch_template.h 6e8b25e426 Refactor il y a 2 mois
mainloop_bwd_sm90_tma_gmma_ws.hpp ea7a98f15d Fix backward with softcap il y a 2 mois
mainloop_fwd_sm90_tma_gmma_ws.hpp 6e8b25e426 Refactor il y a 2 mois
named_barrier.hpp 6e8b25e426 Refactor il y a 2 mois
setup.py 6e8b25e426 Refactor il y a 2 mois
softmax.h 6e8b25e426 Refactor il y a 2 mois
static_switch.h 6e8b25e426 Refactor il y a 2 mois
test_flash_attn.py ea7a98f15d Fix backward with softcap il y a 2 mois
tile_scheduler.hpp 6e8b25e426 Refactor il y a 2 mois
utils.h 6e8b25e426 Refactor il y a 2 mois