Skip to content

Commit

Permalink
update flash_attn
Browse files Browse the repository at this point in the history
  • Loading branch information
@xzyaoi
xzyaoi committed Sep 27, 2024
1 parent 8135d3a commit 605803c
Showing 1 changed file with 1 addition and 1 deletion.
2 changes: 1 addition & 1 deletion 3rdparty/flash-attention
Submodule flash-attention updated 40 files
+1 −1 csrc/composable_kernel
+50 −111 csrc/flash_attn/flash_api.cpp
+43 −20 csrc/flash_attn_ck/flash_api.cpp
+34 −0 csrc/flash_attn_ck/flash_common.cpp
+39 −1 csrc/flash_attn_ck/flash_common.hpp
+88 −65 csrc/flash_attn_ck/mha_bwd.cpp
+34 −56 csrc/flash_attn_ck/mha_fwd.cpp
+568 −0 csrc/flash_attn_ck/mha_fwd_kvcache.cpp
+91 −68 csrc/flash_attn_ck/mha_varlen_bwd.cpp
+29 −52 csrc/flash_attn_ck/mha_varlen_fwd.cpp
+12 −7 flash_attn/bert_padding.py
+400 −117 flash_attn/flash_attn_interface.py
+1 −1 flash_attn/flash_blocksparse_attention.py
+61 −14 flash_attn/layers/rotary.py
+3 −2 flash_attn/losses/cross_entropy.py
+3 −3 flash_attn/models/bert.py
+33 −21 flash_attn/ops/triton/cross_entropy.py
+1 −1 flash_attn/ops/triton/rotary.py
+7 −3 hopper/epilogue_bwd_sm90_tma.hpp
+4 −1 hopper/flash.h
+73 −22 hopper/flash_api.cpp
+69 −8 hopper/flash_attn_interface.py
+10 −9 hopper/flash_bwd_kernel.h
+24 −13 hopper/flash_bwd_launch_template.h
+5 −2 hopper/flash_bwd_postprocess_kernel.h
+5 −2 hopper/flash_bwd_preprocess_kernel.h
+14 −10 hopper/flash_fwd_kernel.h
+40 −31 hopper/flash_fwd_launch_template.h
+68 −15 hopper/mainloop_bwd_sm90_tma_gmma_ws.hpp
+75 −20 hopper/mainloop_fwd_sm90_tma_gmma_ws.hpp
+8 −1 hopper/setup.py
+56 −14 hopper/test_flash_attn.py
+1 −1 hopper/tile_scheduler.hpp
+32 −14 setup.py
+21 −6 tests/losses/test_cross_entropy.py
+21 −5 tests/losses/test_cross_entropy_parallel.py
+3 −3 tests/test_flash_attn.py
+875 −9 tests/test_flash_attn_ck.py
+25 −9 tests/test_rotary.py
+15 −4 tests/test_util.py

0 comments on commit 605803c

Please sign in to comment.