diff --git a/llama.cpp b/llama.cpp
index a7d53d890..ff467c575 100644
--- a/llama.cpp
+++ b/llama.cpp
@@ -9036,7 +9036,7 @@ static int llama_decode_internal(
     //llama_synchronize(&lctx);
 
     // decide if we need to defrag the kv cache
-    if (cparams.defrag_thold >= 0.0f) {
+    if (cparams.causal_attn && cparams.defrag_thold >= 0.0f) {
         const float fragmentation = kv_self.n >= 128 ? 1.0f - float(kv_self.used)/float(kv_self.n) : 0.0f;
 
         // queue defragmentation for next llama_kv_cache_update