diff --git a/gemma/flash_attention.cc b/gemma/flash_attention.cc
index c56c036f..2129a63f 100644
--- a/gemma/flash_attention.cc
+++ b/gemma/flash_attention.cc
@@ -57,7 +57,7 @@ namespace gcpp {
 namespace HWY_NAMESPACE {
 
 static constexpr size_t kNFx8HTileSize = 8;
-
+static constexpr float kNegInf = -std::numeric_limits<float>::max() / 64.0f;
 // Transposes q into q_t.
 // Both are 4D tensors stuffed into a 2-D MatPtrT.
 // q has shape [batch, qbatch][head, qkv_dim].
@@ -467,7 +467,7 @@ static void HWY_INLINE FlashAttentionTileStepAndApplySoftCap(
   const DF4 df4;
   using VF4 = hn::Vec<DF4>;
   static_assert(kNumQueries >= 1 && kNumQueries <= 4);
-  VF4 new_max = hn::Set(df4, -std::numeric_limits<float>::max() / 2.0f);
+  VF4 new_max = hn::Set(df4, kNegInf);
   VF max_0, max_1, max_2, max_3 = hn::Zero(df);
   max_0 = hn::Max(x_0_p0, x_0_p1);
   if constexpr (kNumQueries >= 2) {
@@ -490,29 +490,29 @@ static void HWY_INLINE FlashAttentionTileStepAndApplySoftCap(
     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)));
   }
-  VF4 old_max_vf = hn::Set(df4, -std::numeric_limits<float>::max() / 2.0f);
+  VF4 old_max_vf = hn::Set(df4, kNegInf);
   old_max_vf = hn::LoadU(df4, old_max);
   new_max = hn::Max(new_max, old_max_vf);
+  auto changed_max = hn::Gt(new_max, hn::Set(df4, kNegInf));
   // TODO figure out what was wrong with broadcasts and change to that.
-  HWY_ALIGN float tmp_max[4];
-  hn::Store(new_max, df4, tmp_max);
+  hn::StoreU(new_max, df4, old_max);
   if constexpr (kNumQueries >= 1) {
-    const VF new_max_0 = hn::Set(df, tmp_max[0]);
+    const VF new_max_0 = hn::Set(df, old_max[0]);
     x_0_p0 = hn::Exp(df, hn::Sub(x_0_p0, new_max_0));
     x_0_p1 = hn::Exp(df, hn::Sub(x_0_p1, new_max_0));
   }
   if constexpr (kNumQueries >= 2) {
-    const VF new_max_0 = hn::Set(df, tmp_max[1]);
+    const VF new_max_0 = hn::Set(df, old_max[1]);
     x_1_p0 = hn::Exp(df, hn::Sub(x_1_p0, new_max_0));
     x_1_p1 = hn::Exp(df, hn::Sub(x_1_p1, new_max_0));
   }
   if constexpr (kNumQueries >= 3) {
-    const VF new_max_0 = hn::Set(df, tmp_max[2]);
+    const VF new_max_0 = hn::Set(df, old_max[2]);
     x_2_p0 = hn::Exp(df, hn::Sub(x_2_p0, new_max_0));
     x_2_p1 = hn::Exp(df, hn::Sub(x_2_p1, new_max_0));
   }
   if constexpr (kNumQueries >= 4) {
-    const VF new_max_0 = hn::Set(df, tmp_max[3]);
+    const VF new_max_0 = hn::Set(df, old_max[3]);
     x_3_p0 = hn::Exp(df, hn::Sub(x_3_p0, new_max_0));
     x_3_p1 = hn::Exp(df, hn::Sub(x_3_p1, new_max_0));
   }
@@ -520,8 +520,6 @@ static void HWY_INLINE FlashAttentionTileStepAndApplySoftCap(
   old_d_vf = hn::LoadU(df4, old_d);
   VF4 scale = hn::Mul(old_d_vf, hn::Exp(df4, hn::Sub(old_max_vf, new_max)));
 
-  hn::StoreU(new_max, df4, old_max);
-
   VF4 x_sum = hn::Zero(df4);
   if constexpr (kNumQueries == 1) {
     x_sum = hn::Set(df4, hn::ReduceSum(df, x_0_p0) + hn::ReduceSum(df, x_0_p1));
@@ -539,12 +537,12 @@ static void HWY_INLINE FlashAttentionTileStepAndApplySoftCap(
   const VF4 zero4 = hn::Zero(df4);
   const VF4 one_over_d =
       hn::MaskedDivOr(zero4, non_zero_mask, hn::Set(df4, 1.0f), old_d_vf);
-  float tmp_one_over_d[4];
+  HWY_ALIGN float tmp_one_over_d[4];
   hn::Store(one_over_d, df4, tmp_one_over_d);
-  hn::Store(old_d_vf, df4, old_d);
+  hn::BlendedStore(old_d_vf, changed_max, df4, old_d);
   scale = hn::Mul(scale, one_over_d);
-  hn::Store(scale, df4, scales);
-  if (hn::ExtractLane(old_d_vf, 0) > 0.0f) {
+  hn::BlendedStore(scale, changed_max, df4, scales);
+  if (hn::ExtractLane(old_d_vf, 0) > 0.0f && scales[0] != 1.0f) {
     const VF one_over_d_0 = hn::Set(df, tmp_one_over_d[0]);
     x_0_p0 = hn::Mul(x_0_p0, one_over_d_0);
     x_0_p1 = hn::Mul(x_0_p1, one_over_d_0);
@@ -553,7 +551,7 @@ static void HWY_INLINE FlashAttentionTileStepAndApplySoftCap(
     x_0_p1 = zero;
   }
   if constexpr (kNumQueries >= 2) {
-    if (hn::ExtractLane(old_d_vf, 1) > 0.0f) {
+    if (hn::ExtractLane(old_d_vf, 1) > 0.0f && scales[1] != 1.0f) {
       const VF one_over_d_1 = hn::Set(df, tmp_one_over_d[1]);
       x_1_p0 = hn::Mul(x_1_p0, one_over_d_1);
       x_1_p1 = hn::Mul(x_1_p1, one_over_d_1);
@@ -563,7 +561,7 @@ static void HWY_INLINE FlashAttentionTileStepAndApplySoftCap(
     }
   }
   if constexpr (kNumQueries >= 3) {
-    if (hn::ExtractLane(old_d_vf, 2) > 0.0f) {
+    if (hn::ExtractLane(old_d_vf, 2) > 0.0f && scales[2] != 1.0f) {
       const VF one_over_d_2 = hn::Set(df, tmp_one_over_d[2]);
       x_2_p0 = hn::Mul(x_2_p0, one_over_d_2);
       x_2_p1 = hn::Mul(x_2_p1, one_over_d_2);
@@ -573,7 +571,7 @@ static void HWY_INLINE FlashAttentionTileStepAndApplySoftCap(
     }
   }
   if constexpr (kNumQueries >= 4) {
-    if (hn::ExtractLane(old_d_vf, 3) > 0.0f) {
+    if (hn::ExtractLane(old_d_vf, 3) > 0.0f && scales[3] != 1.0f) {
       const VF one_over_d_3 = hn::Set(df, tmp_one_over_d[3]);
       x_3_p0 = hn::Mul(x_3_p0, one_over_d_3);
       x_3_p1 = hn::Mul(x_3_p1, one_over_d_3);