Ying Zhang b5a8931042 doc masking 5 hónapja
..
__init__.py 7f67966cc7 FA3 initial code release 7 hónapja
benchmark_attn.py 3669b25206 bwd benchmark + small fixes (#1129) 7 hónapja
benchmark_flash_attention_fp8.py c92ca63268 FA3 FP8 qkv descales + restore max offset for h128 causal + added sync for producer WG (#1173) 6 hónapja
block_info.h 7f67966cc7 FA3 initial code release 7 hónapja
epilogue_bwd_sm90_tma.hpp db80387343 Add seqused_q in fwd / bwd and seqused_k in bwd. 5 hónapja
epilogue_fwd_sm90_tma.hpp a3a257c71d Fix out-of-bound writes for var-seq-len zero-length KVs 6 hónapja
flash.h b5a8931042 doc masking 5 hónapja
flash_api.cpp b5a8931042 doc masking 5 hónapja
flash_attn_interface.py b5a8931042 doc masking 5 hónapja
flash_bwd_hdim128_bf16_sm90.cu bafe253042 [FA3] Bwd 7 hónapja
flash_bwd_hdim128_fp16_sm90.cu 7f67966cc7 FA3 initial code release 7 hónapja
flash_bwd_hdim256_fp16_sm90.cu 7f67966cc7 FA3 initial code release 7 hónapja
flash_bwd_hdim64_bf16_sm90.cu bafe253042 [FA3] Bwd 7 hónapja
flash_bwd_hdim64_fp16_sm90.cu 7f67966cc7 FA3 initial code release 7 hónapja
flash_bwd_hdim96_bf16_sm90.cu bafe253042 [FA3] Bwd 7 hónapja
flash_bwd_hdim96_fp16_sm90.cu bafe253042 [FA3] Bwd 7 hónapja
flash_bwd_kernel.h 1c9717d699 address comments 5 hónapja
flash_bwd_launch_template.h 1c9717d699 address comments 5 hónapja
flash_bwd_postprocess_kernel.h db80387343 Add seqused_q in fwd / bwd and seqused_k in bwd. 5 hónapja
flash_bwd_preprocess_kernel.h db80387343 Add seqused_q in fwd / bwd and seqused_k in bwd. 5 hónapja
flash_fwd_hdim128_bf16_sm90.cu 74b0761ff7 [FA3] BF16 forward 7 hónapja
flash_fwd_hdim128_e4m3_sm90.cu 5018ac6ac5 Fp8 kernel with "in-kernel" transpose of V in producer (#1100) 7 hónapja
flash_fwd_hdim128_fp16_sm90.cu 7f67966cc7 FA3 initial code release 7 hónapja
flash_fwd_hdim256_bf16_sm90.cu 74b0761ff7 [FA3] BF16 forward 7 hónapja
flash_fwd_hdim256_e4m3_sm90.cu 5018ac6ac5 Fp8 kernel with "in-kernel" transpose of V in producer (#1100) 7 hónapja
flash_fwd_hdim256_fp16_sm90.cu 7f67966cc7 FA3 initial code release 7 hónapja
flash_fwd_hdim64_bf16_sm90.cu 74b0761ff7 [FA3] BF16 forward 7 hónapja
flash_fwd_hdim64_e4m3_sm90.cu 5018ac6ac5 Fp8 kernel with "in-kernel" transpose of V in producer (#1100) 7 hónapja
flash_fwd_hdim64_fp16_sm90.cu 7f67966cc7 FA3 initial code release 7 hónapja
flash_fwd_kernel.h b5a8931042 doc masking 5 hónapja
flash_fwd_launch_template.h b5a8931042 doc masking 5 hónapja
kernel_traits.h b5a8931042 doc masking 5 hónapja
mainloop_bwd_sm90_tma_gmma_ws.hpp 1c9717d699 address comments 5 hónapja
mainloop_fwd_sm90_tma_gmma_ws.hpp dff976a84a fixes 5 hónapja
named_barrier.hpp bafe253042 [FA3] Bwd 7 hónapja
seq_len.h b5a8931042 doc masking 5 hónapja
setup.py 8476986721 Fix FAv3 compilation with MSVC (#1240) 5 hónapja
softmax.h c92ca63268 FA3 FP8 qkv descales + restore max offset for h128 causal + added sync for producer WG (#1173) 6 hónapja
static_switch.h dfe1a59e4b Add var-seq-len to FA3 fp16 / bf16 fwd (#1072) 7 hónapja
test_flash_attn.py b5a8931042 doc masking 5 hónapja
tile_scheduler.hpp b5a8931042 doc masking 5 hónapja
tile_scheduler_bwd.hpp bafe253042 [FA3] Bwd 7 hónapja
utils.h bafe253042 [FA3] Bwd 7 hónapja