diff --git a/ggml/include/ggml.h b/ggml/include/ggml.h
index 250d924ed..b3a6c8adb 100644
--- a/ggml/include/ggml.h
+++ b/ggml/include/ggml.h
@@ -1269,6 +1269,12 @@ extern "C" {
             struct ggml_tensor  * a,
             struct ggml_tensor  * b,
             int                   dim);
+    GGML_API struct ggml_tensor * ggml_concat_inplace(
+            struct ggml_context * ctx,
+            struct ggml_tensor  * a,
+            struct ggml_tensor  * b,
+            struct ggml_tensor  * result,
+            int                   dim);
 
     GGML_API struct ggml_tensor * ggml_abs(
             struct ggml_context * ctx,
diff --git a/ggml/src/ggml.c b/ggml/src/ggml.c
index b7db11d9f..6a551cc3f 100644
--- a/ggml/src/ggml.c
+++ b/ggml/src/ggml.c
@@ -6944,6 +6944,40 @@ struct ggml_tensor * ggml_concat(
     return result;
 }
 
+struct ggml_tensor * ggml_concat_inplace(
+    struct ggml_context * ctx,
+    struct ggml_tensor * a,
+    struct ggml_tensor * b,
+    struct ggml_tensor * result_in,
+    int dim) {
+    GGML_ASSERT(dim >= 0 && dim < GGML_MAX_DIMS);
+    if (!result_in) {
+        return ggml_concat(ctx, a, b, dim);
+    }
+
+    for (int d = 0; d < GGML_MAX_DIMS; ++d) {
+        int64_t ne;
+        if (d == dim) {
+            ne = a->ne[d] + b->ne[d];
+        } else {
+            GGML_ASSERT(a->ne[d] == b->ne[d]);
+            ne = a->ne[d];
+        }
+        GGML_ASSERT(ne == result_in->ne[d]);
+    }
+
+    struct ggml_tensor * result = ggml_view_tensor(ctx, result_in);
+
+    ggml_set_op_params_i32(result, 0, dim);
+
+    result->op = GGML_OP_CONCAT;
+    result->grad = NULL;
+    result->src[0] = a;
+    result->src[1] = b;
+
+    return result;
+}
+
 // ggml_abs
 
 struct ggml_tensor * ggml_abs(
diff --git a/src/llama-delta-net.cpp b/src/llama-delta-net.cpp
index 81c510209..92f3748ff 100644
--- a/src/llama-delta-net.cpp
+++ b/src/llama-delta-net.cpp
@@ -405,10 +405,7 @@ ggml_tensor * delta_net::build_qkv(ggml_context * ctx0, ggml_tensor * state_stor
     cb(new_conv_states_cont, "new_conv_states_cont", il);
     ggml_tensor * new_conv_flat = ggml_reshape_2d(ctx0, new_conv_states_cont, conv_state_dim, 1);
     ggml_tensor * new_ssm_flat  = ggml_reshape_2d(ctx0, new_state, ssm_state_dim, 1);
-    ggml_tensor * new_state_flat = ggml_concat(ctx0, new_conv_flat, new_ssm_flat, 0);
-    cb(new_state_flat, "new_state_flat", il);
-
-    auto state_cpy = ggml_cpy(ctx0, new_state_flat, state_dst);
+    auto state_cpy = ggml_concat_inplace(ctx0, new_conv_flat, new_ssm_flat, state_dst, 0);
     cb(state_cpy, "state_cpy", il);
     ggml_build_forward_expand(gf, state_cpy);