From bf1d0561254fb8a45be2d3e9897a8dd917301839 Mon Sep 17 00:00:00 2001
From: Iwan Kawrakow <iwan.kawrakow@gmail.com>
Date: Sun, 9 Feb 2025 09:24:52 +0200
Subject: [PATCH] Make sure we do have wk_b and wv_b before enabling MLA

---
 src/llama.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/llama.cpp b/src/llama.cpp
index 1d7ecbde..17d25733 100644
--- a/src/llama.cpp
+++ b/src/llama.cpp
@@ -3175,7 +3175,7 @@ static bool llama_kv_cache_init(
         struct ggml_context * ctx = offload ? ctx_map.at(model.buft_layer[i].buft) : cache.ctxs.front();
         ggml_tensor * k;
         ggml_tensor * v;
-        if (cparams.mla_attn) {
+        if (cparams.mla_attn && model.layers[i].wk_b && model.layers[i].wv_b) {
                k = ggml_new_tensor_1d(ctx, type_k, 1);
                v = ggml_new_tensor_1d(ctx, type_v, 1);
         }