Update multihead_attention.cu

microsoft · Jan 23, 2024 · 63efcf4 · 63efcf4
1 parent 67be406
commit 63efcf4
Showing 1 changed file with 2 additions and 1 deletion.
diff --git a/onnxruntime/contrib_ops/rocm/bert/multihead_attention.cu b/onnxruntime/contrib_ops/rocm/bert/multihead_attention.cu
@@ -123,7 +123,8 @@ Status MultiHeadAttention<T>::ComputeInternal(OpKernelContext* context) const {
           key_padding_mask, relative_position_bias,
           past_key, past_value, past_seq_len,
           &attn,
-          num_heads_, is_unidirectional_, mask_filter_value_, scale_,
+          num_heads_, false, /*is_unidirectional_*/ 
+          mask_filter_value_, scale_,
           past_present_share_buffer_, false, device_prop.maxThreadsPerBlock));
 
   if (attn_type_ == kDecoderMaskedMultiHeadAttention && attn.sequence_length != 1) {