CUDA: fix Gemma E4B MTP FlashAttention (#25148)

* CUDA: fix Gemma E4B MTP FlashAttention * remove unused template declaration
2026-06-30 14:06:54 +02:00
parent f708a5b2ca
commit e495d1e748
8 changed files with 23 additions and 11 deletions
@@ -99,12 +99,12 @@ static void ggml_cuda_flash_attn_ext_mma_f16_switch_ncols2(ggml_backend_cuda_con
        return;
    }

-    if constexpr (DKQ <= 256) {
-        if (use_gqa_opt && gqa_ratio > 1) {
-            ggml_cuda_flash_attn_ext_mma_f16_switch_ncols1<DKQ, DV, 2>(ctx, dst);
-            return;
-        }
+    if (use_gqa_opt && gqa_ratio > 1) {
+        ggml_cuda_flash_attn_ext_mma_f16_switch_ncols1<DKQ, DV, 2>(ctx, dst);
+        return;
+    }

+    if constexpr (DKQ <= 256) {
        ggml_cuda_flash_attn_ext_mma_f16_switch_ncols1<DKQ, DV, 1>(ctx, dst);
    } else {
        GGML_ABORT("fatal error");