From 84d8eb419d6d9876d942cd953bf877fe3a52c4cf Mon Sep 17 00:00:00 2001 From: Dani Ferreira Franco Moura Date: Thu, 26 Feb 2026 06:54:42 -0800 Subject: [PATCH] internal change PiperOrigin-RevId: 875702115 --- gemma/flash_attention.cc | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/gemma/flash_attention.cc b/gemma/flash_attention.cc index ebe8ee18..70a55b8f 100644 --- a/gemma/flash_attention.cc +++ b/gemma/flash_attention.cc @@ -972,8 +972,8 @@ static HWY_INLINE void QDotKTilexUpTo8TransposedKDoubleWidth( for (size_t i = 0; i < qkv_dim; ++i) { VQ_T k_vec1, k_vec2; if constexpr (HWY_TARGET == HWY_AVX2) { - hwy::Prefetch(k_transposed_span.ptr + (i + 3) * gcpp::KVCache::kTileSize); - hwy::Prefetch(k_transposed_span.ptr + (i + 4) * gcpp::KVCache::kTileSize); + hwy::Prefetch(k_transposed_span.ptr + + (i + 20) * gcpp::KVCache::kTileSize); } Decompress2(df, k_transposed_span, i * gcpp::KVCache::kTileSize, k_vec1, k_vec2);