diff --git a/extension/llm/custom_ops/op_sdpa_impl.h b/extension/llm/custom_ops/op_sdpa_impl.h index 9c479569505..dbf91b7d223 100644 --- a/extension/llm/custom_ops/op_sdpa_impl.h +++ b/extension/llm/custom_ops/op_sdpa_impl.h @@ -443,9 +443,7 @@ _exp_reduce_sum_fusion_kernel(T1* a, const int& size, T2* out, T1& val) { for (int i = 0; i < vec_size * (size / vec_size); i += vec_size) { auto tmp0 = vec::VectorizedN::loadu(a + i); auto tmp1 = tmp0 - vec_max; - // Replace with exp_u20 later - // auto tmp2 = tmp1.exp_u20(); - auto tmp2 = tmp1.exp(); + auto tmp2 = tmp1.exp_u20(); vec_tmp_sum = vec_tmp_sum + tmp2; tmp2.store(out + i); }