Jay Shah c272619bf5 disable clusters 1 kuukausi sitten
..
__init__.py 7f67966cc7 FA3 initial code release 5 kuukautta sitten
benchmark_attn.py 3669b25206 bwd benchmark + small fixes (#1129) 4 kuukautta sitten
benchmark_flash_attention_fp8.py 483b26e541 add working fp8 varlen 1 kuukausi sitten
benchmark_split_kv.py a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
combine.h 478ee666cc Make namespace comment consistent (#1305) 1 kuukausi sitten
epilogue_bwd_sm90_tma.hpp db80387343 Add seqused_q in fwd / bwd and seqused_k in bwd. 3 kuukautta sitten
epilogue_fwd_sm90_tma.hpp bb169e2a54 revert to single tile scheduler 1 kuukausi sitten
flash.h a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_api.cpp bb169e2a54 revert to single tile scheduler 1 kuukausi sitten
flash_attn_interface.py 483b26e541 add working fp8 varlen 1 kuukausi sitten
flash_bwd_hdim128_bf16_sm90.cu bafe253042 [FA3] Bwd 4 kuukautta sitten
flash_bwd_hdim128_fp16_sm90.cu 7f67966cc7 FA3 initial code release 5 kuukautta sitten
flash_bwd_hdim256_fp16_sm90.cu 7f67966cc7 FA3 initial code release 5 kuukautta sitten
flash_bwd_hdim64_bf16_sm90.cu bafe253042 [FA3] Bwd 4 kuukautta sitten
flash_bwd_hdim64_fp16_sm90.cu 7f67966cc7 FA3 initial code release 5 kuukautta sitten
flash_bwd_hdim96_bf16_sm90.cu bafe253042 [FA3] Bwd 4 kuukautta sitten
flash_bwd_hdim96_fp16_sm90.cu bafe253042 [FA3] Bwd 4 kuukautta sitten
flash_bwd_kernel.h 1c9717d699 address comments 2 kuukautta sitten
flash_bwd_launch_template.h 1c9717d699 address comments 2 kuukautta sitten
flash_bwd_postprocess_kernel.h db80387343 Add seqused_q in fwd / bwd and seqused_k in bwd. 3 kuukautta sitten
flash_bwd_preprocess_kernel.h db80387343 Add seqused_q in fwd / bwd and seqused_k in bwd. 3 kuukautta sitten
flash_fwd_hdim128_bf16_gqa16_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim128_bf16_gqa2_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim128_bf16_gqa32_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim128_bf16_gqa4_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim128_bf16_gqa8_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim128_bf16_sm90.cu 74b0761ff7 [FA3] BF16 forward 5 kuukautta sitten
flash_fwd_hdim128_e4m3_gqa16_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim128_e4m3_gqa2_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim128_e4m3_gqa32_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim128_e4m3_gqa4_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim128_e4m3_gqa8_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim128_e4m3_sm90.cu 5018ac6ac5 Fp8 kernel with "in-kernel" transpose of V in producer (#1100) 4 kuukautta sitten
flash_fwd_hdim128_fp16_gqa16_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim128_fp16_gqa2_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim128_fp16_gqa32_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim128_fp16_gqa4_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim128_fp16_gqa8_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim128_fp16_sm90.cu 7f67966cc7 FA3 initial code release 5 kuukautta sitten
flash_fwd_hdim256_bf16_gqa16_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim256_bf16_gqa2_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim256_bf16_gqa32_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim256_bf16_gqa4_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim256_bf16_gqa8_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim256_bf16_sm90.cu 74b0761ff7 [FA3] BF16 forward 5 kuukautta sitten
flash_fwd_hdim256_e4m3_gqa16_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim256_e4m3_gqa2_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim256_e4m3_gqa32_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim256_e4m3_gqa4_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim256_e4m3_gqa8_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim256_e4m3_sm90.cu 5018ac6ac5 Fp8 kernel with "in-kernel" transpose of V in producer (#1100) 4 kuukautta sitten
flash_fwd_hdim256_fp16_gqa16_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim256_fp16_gqa2_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim256_fp16_gqa32_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim256_fp16_gqa4_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim256_fp16_gqa8_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim256_fp16_sm90.cu 7f67966cc7 FA3 initial code release 5 kuukautta sitten
flash_fwd_hdim64_bf16_gqa16_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim64_bf16_gqa2_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim64_bf16_gqa32_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim64_bf16_gqa4_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim64_bf16_gqa8_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim64_bf16_sm90.cu 74b0761ff7 [FA3] BF16 forward 5 kuukautta sitten
flash_fwd_hdim64_e4m3_gqa16_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim64_e4m3_gqa2_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim64_e4m3_gqa32_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim64_e4m3_gqa4_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim64_e4m3_gqa8_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim64_e4m3_sm90.cu 5018ac6ac5 Fp8 kernel with "in-kernel" transpose of V in producer (#1100) 4 kuukautta sitten
flash_fwd_hdim64_fp16_gqa16_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim64_fp16_gqa2_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim64_fp16_gqa32_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim64_fp16_gqa4_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim64_fp16_gqa8_sm90.cu a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
flash_fwd_hdim64_fp16_sm90.cu 7f67966cc7 FA3 initial code release 5 kuukautta sitten
flash_fwd_kernel.h c272619bf5 disable clusters 1 kuukausi sitten
flash_fwd_launch_template.h c272619bf5 disable clusters 1 kuukausi sitten
kernel_traits.h bb169e2a54 revert to single tile scheduler 1 kuukausi sitten
mainloop_bwd_sm90_tma_gmma_ws.hpp 1c9717d699 address comments 2 kuukautta sitten
mainloop_fwd_sm90_tma_gmma_ws.hpp bb169e2a54 revert to single tile scheduler 1 kuukausi sitten
named_barrier.hpp bb169e2a54 revert to single tile scheduler 1 kuukausi sitten
seq_len.h 483b26e541 add working fp8 varlen 1 kuukausi sitten
setup.py a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
softmax.h 478ee666cc Make namespace comment consistent (#1305) 1 kuukausi sitten
static_switch.h a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
test_attn_kvcache.py a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
test_flash_attn.py c272619bf5 disable clusters 1 kuukausi sitten
test_kvcache.py a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) 2 kuukautta sitten
tile_scheduler.hpp 478ee666cc Make namespace comment consistent (#1305) 1 kuukausi sitten
tile_scheduler_bwd.hpp 478ee666cc Make namespace comment consistent (#1305) 1 kuukausi sitten
utils.h 478ee666cc Make namespace comment consistent (#1305) 1 kuukausi sitten