diff --git a/src/layers/flash_attention.cc b/src/layers/flash_attention.cc index 45b3a49ed..26881f5fa 100644 --- a/src/layers/flash_attention.cc +++ b/src/layers/flash_attention.cc @@ -18,6 +18,7 @@ namespace ctranslate2 { void FlashMultiHeadAttention::operator()(const StorageView& queries, const StorageView&, const StorageView* values_lengths, + const StorageView* values_offsets, StorageView& output, StorageView* cached_keys, StorageView* cached_values, diff --git a/src/layers/wav2vec2bert.cc b/src/layers/wav2vec2bert.cc index 7ac3620e5..3b5d7c87d 100644 --- a/src/layers/wav2vec2bert.cc +++ b/src/layers/wav2vec2bert.cc @@ -51,6 +51,7 @@ namespace ctranslate2 { (*_self_attention)(buffer1, buffer1, nullptr, + nullptr, buffer2, nullptr, nullptr, @@ -147,6 +148,7 @@ namespace ctranslate2 { (*_self_attention)(buffer2, buffer2, nullptr, + nullptr, buffer1, nullptr, nullptr,