diff --git a/ggml-kompute.cpp b/ggml-kompute.cpp
index d7e7984c1..3f033d58b 100644
--- a/ggml-kompute.cpp
+++ b/ggml-kompute.cpp
@@ -1559,12 +1559,19 @@ static void ggml_vk_graph_compute(struct ggml_kompute_context * ctx, struct ggml
                 case GGML_OP_SOFT_MAX:
                     {
                         float scale;
-                        memcpy(&scale, dst->op_params, sizeof(float));
+                        float max_bias;
+
+                        memcpy(&scale,    (float *)dst->op_params + 0, sizeof(float));
+                        memcpy(&max_bias, (float *)dst->op_params + 1, sizeof(float));
 
 #pragma message("TODO: add ggml_vk_soft_max() F16 src1 support")
 #pragma message("ref:  https://github.com/ggerganov/llama.cpp/pull/5021")
                         GGML_ASSERT(!src1 || src1t == GGML_TYPE_F32);
 
+#pragma message("TODO: add ALiBi support")
+#pragma message("ref:  https://github.com/ggerganov/llama.cpp/pull/7192")
+                        GGML_ASSERT(max_bias == 0.0f);
+
                         ggml_vk_soft_max(seq, id_src0, id_src1, id_dst, off_src0, off_src1, off_dst, ne00, ne01, ne02, ne03, scale);
                     } break;
                 case GGML_OP_DIAG_MASK_INF:
diff --git a/ggml-vulkan.cpp b/ggml-vulkan.cpp
index 4d32c2f27..b9449be03 100644
--- a/ggml-vulkan.cpp
+++ b/ggml-vulkan.cpp
@@ -4285,6 +4285,9 @@ static void ggml_vk_soft_max(ggml_backend_vk_context * ctx, vk_context * subctx,
     const float m0 = powf(2.0f, -(max_bias       ) / n_head_log2);
     const float m1 = powf(2.0f, -(max_bias / 2.0f) / n_head_log2);
 
+#pragma message("TODO: src2 is no longer used in soft_max - should be removed and ALiBi calculation should be updated")
+#pragma message("ref:  https://github.com/ggerganov/llama.cpp/pull/7192")
+
     ggml_vk_op_f32<vk_op_soft_max_push_constants>(ctx, subctx, src0, src1, src2, dst, GGML_OP_SOFT_MAX, {
         ncols,
         src1 != nullptr ? nrows_y : (uint32_t)0,