diff --git a/ggml/src/ggml-cuda/fattn-new-mma.cu b/ggml/src/ggml-cuda/fattn-new-mma.cu
index 8400a725..52a14639 100644
--- a/ggml/src/ggml-cuda/fattn-new-mma.cu
+++ b/ggml/src/ggml-cuda/fattn-new-mma.cu
@@ -2148,11 +2148,15 @@ void ggml_cuda_flash_attn_ext_mma_new(ggml_backend_cuda_context & ctx, ggml_tens
     GGML_ASSERT(Q->ne[2] % K->ne[2] == 0);
     const int gqa_ratio = Q->ne[2] / K->ne[2];
 
-    if (K->ne[0] == 128 && gqa_ratio == 12) {
+    if (K->ne[0] == 128 && (gqa_ratio == 12 || gqa_ratio == 6)) {
         GGML_ASSERT(Q->ne[0] == 128 && V->ne[0] == 128);
         //GGML_ASSERT(Q->ne[1] <= 4);
         //ggml_cuda_flash_attn_ext_mma_f16_switch_ncols1<128, 128, 16>(ctx, dst);
-        ggml_cuda_flash_attn_ext_mma_f16_case<128, 128, 1, 16>(ctx, dst);
+        if (gqa_ratio == 12) {
+            ggml_cuda_flash_attn_ext_mma_f16_case<128, 128, 1, 16>(ctx, dst);
+        } else {
+            ggml_cuda_flash_attn_ext_mma_f16_case<128, 128, 1, 8>(ctx, dst);
+        }
         return;
     }
     //if (K->ne[0] == 64 && V->ne[0] == 64) {
diff --git a/ggml/src/ggml-cuda/fattn.cu b/ggml/src/ggml-cuda/fattn.cu
index 6f1c766c..267968b0 100644
--- a/ggml/src/ggml-cuda/fattn.cu
+++ b/ggml/src/ggml-cuda/fattn.cu
@@ -90,7 +90,7 @@ void ggml_cuda_flash_attn_ext(ggml_backend_cuda_context & ctx, ggml_tensor * dst
     }
 
     if (new_mma_available(cc) && K->ne[0] == 128 && V->ne[0] == 128 && Q->ne[0] == 128 && Q->ne[1] == 1 &&
-            Q->ne[2] / K->ne[2] == 12) {
+            (Q->ne[2] / K->ne[2] == 12 || Q->ne[2] / K->ne[2] == 6)) {
         ggml_cuda_flash_attn_ext_mma_new(ctx, dst);
         return;
     }