Skip to content

Replace Tanh with FastTanh in flash_attention #900

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
copybara-service wants to merge 1 commit into
base: dev
Choose a base branch
from
+11 −11
Open

Replace Tanh with FastTanh in flash_attention #900

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
20 changes: 10 additions & 10 deletions gemma/flash_attention.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -591,7 +591,7 @@ static HWY_INLINE void FlashAttentionTileStepAndApplySoftCap4(
if (att_cap > 0.0f) {
VF4 cap = hn::Set(df4, att_cap);
VF4 one_over_cap = hn::Set(df4, one_over_att_cap);
new_max = hn::Mul(cap, hn::Tanh(df4, hn::Mul(new_max, one_over_cap)));
new_max = hn::Mul(cap, hn::FastTanh(df4, hn::Mul(new_max, one_over_cap)));
}
VF4 local_max = new_max;
VF4 old_max_vf = hn::Set(df4, kNegInf);
Expand Down Expand Up @@ -733,7 +733,7 @@ static HWY_INLINE void FlashAttentionTileStepAndApplySoftCap8(
if (att_cap > 0.0f) {
VF8 cap = hn::Set(df8, att_cap);
VF8 one_over_cap = hn::Set(df8, one_over_att_cap);
new_max = hn::Mul(cap, hn::Tanh(df8, hn::Mul(new_max, one_over_cap)));
new_max = hn::Mul(cap, hn::FastTanh(df8, hn::Mul(new_max, one_over_cap)));
}
VF8 local_max = new_max;
VF8 old_max_vf = hn::Set(df8, kNegInf);
Expand Down Expand Up @@ -1309,27 +1309,27 @@ static HWY_INLINE void ApplySoftCap(DF df, float att_cap, float one_over_cap,
if (att_cap > 0.0f) {
VF cap = hn::Set(df, att_cap);
VF one_over_cap_vec = hn::Set(df, one_over_cap);
x0 = hn::Mul(cap, hn::CallTanh(df, hn::Mul(x0, one_over_cap_vec)));
x0 = hn::Mul(cap, hn::CallFastTanh(df, hn::Mul(x0, one_over_cap_vec)));
if constexpr (kVTileSize >= 2) {
x1 = hn::Mul(cap, hn::CallTanh(df, hn::Mul(x1, one_over_cap_vec)));
x1 = hn::Mul(cap, hn::CallFastTanh(df, hn::Mul(x1, one_over_cap_vec)));
}
if constexpr (kVTileSize >= 3) {
x2 = hn::Mul(cap, hn::CallTanh(df, hn::Mul(x2, one_over_cap_vec)));
x2 = hn::Mul(cap, hn::CallFastTanh(df, hn::Mul(x2, one_over_cap_vec)));
}
if constexpr (kVTileSize >= 4) {
x3 = hn::Mul(cap, hn::CallTanh(df, hn::Mul(x3, one_over_cap_vec)));
x3 = hn::Mul(cap, hn::CallFastTanh(df, hn::Mul(x3, one_over_cap_vec)));
}
if constexpr (kVTileSize >= 5) {
x4 = hn::Mul(cap, hn::CallTanh(df, hn::Mul(x4, one_over_cap_vec)));
x4 = hn::Mul(cap, hn::CallFastTanh(df, hn::Mul(x4, one_over_cap_vec)));
}
if constexpr (kVTileSize >= 6) {
x5 = hn::Mul(cap, hn::CallTanh(df, hn::Mul(x5, one_over_cap_vec)));
x5 = hn::Mul(cap, hn::CallFastTanh(df, hn::Mul(x5, one_over_cap_vec)));
}
if constexpr (kVTileSize >= 7) {
x6 = hn::Mul(cap, hn::CallTanh(df, hn::Mul(x6, one_over_cap_vec)));
x6 = hn::Mul(cap, hn::CallFastTanh(df, hn::Mul(x6, one_over_cap_vec)));
}
if constexpr (kVTileSize >= 8) {
x7 = hn::Mul(cap, hn::CallTanh(df, hn::Mul(x7, one_over_cap_vec)));
x7 = hn::Mul(cap, hn::CallFastTanh(df, hn::Mul(x7, one_over_cap_vec)));
}
}
}
Expand Down
2 changes: 1 addition & 1 deletion gemma/flash_attention_test.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -257,7 +257,7 @@ void AssertClose(const MatPtrT<float>& a, const MatPtrT<float>& b) {
if (rel_abs_delta > 0.0f) {
rel_abs_delta /= std::max(std::abs(a_row[c]), std::abs(b_row[c]));
}
EXPECT_LT(rel_abs_delta, 1e-3)
EXPECT_LT(rel_abs_delta, 3e-3)
<< "a[" << r << "," << c << "]=" << a_row[c] << ", b[" << r << ","
<< c << "]=" << b_row[c];
}
Expand Down
Loading