diff --git a/server/Makefile-flash-att-v2 b/server/Makefile-flash-att-v2 index 14c7829f..03f30f8a 100644 --- a/server/Makefile-flash-att-v2 +++ b/server/Makefile-flash-att-v2 @@ -1,6 +1,10 @@ flash_att_v2_commit_cuda := v2.5.9.post1 flash_att_v2_commit_rocm := 2554f490101742ccdc56620a938f847f61754be6 +build-flash-attention-v2-cuda: + pip install -U packaging wheel + pip install flash-attn==$(flash_att_v2_commit_cuda) + install-flash-attention-v2-cuda: pip install -U packaging wheel pip install flash-attn==$(flash_att_v2_commit_cuda)