diff --git a/ds4.c b/ds4.c
index ee069b76a..f30021bca 100644
--- a/ds4.c
+++ b/ds4.c
@@ -6911,10 +6911,10 @@ static void layer_grouped_out_one(
         const ds4_model   * model,
         const ds4_layer_weights * layer,
         const float       * heads) {
-    const uint32_t n_groups = 8;
+    const uint32_t n_groups = DS4_N_OUT_GROUP;
     const uint32_t group_heads = DS4_N_HEAD / n_groups;
     const uint32_t group_dim = DS4_N_HEAD_DIM * group_heads;
-    const uint32_t rank = 1024;
+    const uint32_t rank = DS4_N_LORA_O;
 
     float *low = xcalloc((size_t)n_groups * rank, sizeof(low[0]));
 
@@ -6930,10 +6930,10 @@ static void layer_grouped_out_one_decode_scratch(
         const ds4_layer_weights * layer,
         const float            * heads,
         ds4_cpu_decode_scratch * scratch) {
-    const uint32_t n_groups = 8;
+    const uint32_t n_groups = DS4_N_OUT_GROUP;
     const uint32_t group_heads = DS4_N_HEAD / n_groups;
     const uint32_t group_dim = DS4_N_HEAD_DIM * group_heads;
-    const uint32_t rank = 1024;
+    const uint32_t rank = DS4_N_LORA_O;
 
     memset(scratch->attn_low, 0, (size_t)n_groups * rank * sizeof(scratch->attn_low[0]));
     matvec_q8_0_grouped_rows_decode_scratch(scratch->attn_low, model, layer->attn_output_a,
@@ -6947,10 +6947,10 @@ static void layer_grouped_out_batch(
         const ds4_layer_weights * layer,
         const float       * heads,
         uint32_t            n_tok) {
-    const uint32_t n_groups = 8;
+    const uint32_t n_groups = DS4_N_OUT_GROUP;
     const uint32_t group_heads = DS4_N_HEAD / n_groups;
     const uint32_t group_dim = DS4_N_HEAD_DIM * group_heads;
-    const uint32_t rank = 1024;
+    const uint32_t rank = DS4_N_LORA_O;
 
     float *low = xcalloc((size_t)n_tok * n_groups * rank, sizeof(low[0]));