diff --git a/src/models/qwen3next.cpp b/src/models/qwen3next.cpp
index 7434f095ad..efd3c061f0 100644
--- a/src/models/qwen3next.cpp
+++ b/src/models/qwen3next.cpp
@@ -124,11 +124,6 @@ std::pair<ggml_tensor *, ggml_tensor *> llm_build_qwen3next::build_delta_net_chu
 
     GGML_ASSERT(H_k == H_v);  // we did a repeat to make sure this is the case
 
-    const float eps_norm = hparams.f_norm_rms_eps;
-
-    q = ggml_l2_norm(ctx0, q, eps_norm);
-    k = ggml_l2_norm(ctx0, k, eps_norm);
-
     const float scale = 1.0f / sqrtf(S_v);
 
     q = ggml_scale(ctx0, q, scale);
@@ -370,11 +365,6 @@ std::pair<ggml_tensor *, ggml_tensor *> llm_build_qwen3next::build_delta_net_aut
 
     GGML_ASSERT(H_k == H_v);  // we did a repeat to make sure this is the case
 
-    const float eps_norm = hparams.f_norm_rms_eps;
-
-    q = ggml_l2_norm(ctx0, q, eps_norm);
-    k = ggml_l2_norm(ctx0, k, eps_norm);
-
     const float scale = 1.0f / sqrtf(S_k);
 
     q = ggml_scale(ctx0, q, scale);
@@ -715,9 +705,16 @@ ggml_tensor * llm_build_qwen3next::build_layer_attn_linear(
     int64_t nb1_qkv = ggml_row_size(conv_qkv_mix->type, qkv_dim);
 
     // Extract the convolved Q, K, V from conv_output
-    ggml_tensor * q_conv = ggml_view_2d(ctx0, conv_qkv_mix, head_k_dim * num_k_heads, n_seq_tokens * n_seqs, nb1_qkv, 0);
-    ggml_tensor * k_conv = ggml_view_2d(ctx0, conv_qkv_mix, head_k_dim * num_k_heads, n_seq_tokens * n_seqs, nb1_qkv,
-                     head_k_dim * num_k_heads * ggml_element_size(conv_qkv_mix));
+    ggml_tensor * q_conv = ggml_view_4d(ctx0, conv_qkv_mix, head_k_dim, num_k_heads, n_seq_tokens, n_seqs,
+            ggml_row_size(conv_qkv_mix->type, head_k_dim),
+            nb1_qkv,
+            nb1_qkv * n_seq_tokens,
+            0);
+    ggml_tensor * k_conv = ggml_view_4d(ctx0, conv_qkv_mix, head_k_dim, num_k_heads, n_seq_tokens, n_seqs,
+            ggml_row_size(conv_qkv_mix->type, head_k_dim),
+            nb1_qkv,
+            nb1_qkv * n_seq_tokens,
+            head_k_dim * num_k_heads * ggml_element_size(conv_qkv_mix));
 
     ggml_tensor * v_conv = ggml_view_4d(ctx0, conv_qkv_mix, head_v_dim, num_v_heads, n_seq_tokens, n_seqs,
             ggml_row_size(conv_qkv_mix->type, head_v_dim),
@@ -729,9 +726,14 @@ ggml_tensor * llm_build_qwen3next::build_layer_attn_linear(
     cb(k_conv, "k_conv", il);
     cb(v_conv, "v_conv", il);
 
+    const float eps_norm = hparams.f_norm_rms_eps;
+
+    q_conv = ggml_l2_norm(ctx0, q_conv, eps_norm);
+    k_conv = ggml_l2_norm(ctx0, k_conv, eps_norm);
+
     // Unsqueeze them
-    q_conv = ggml_cont_4d(ctx0, q_conv, head_k_dim, num_k_heads, n_seq_tokens, n_seqs);
-    k_conv = ggml_cont_4d(ctx0, k_conv, head_k_dim, num_k_heads, n_seq_tokens, n_seqs);
+  //q_conv = ggml_cont_4d(ctx0, q_conv, head_k_dim, num_k_heads, n_seq_tokens, n_seqs);
+  //k_conv = ggml_cont_4d(ctx0, k_conv, head_k_dim, num_k_heads, n_seq_tokens, n_seqs);
   //v_conv = ggml_cont_4d(ctx0, v_conv, head_v_dim, num_v_heads, n_seq_tokens, n_seqs);
 
     ggml_tensor * state = build_rs(inp, ssm_states_all, hparams.n_embd_s(), n_seqs);
@@ -778,9 +780,9 @@ ggml_tensor * llm_build_qwen3next::build_layer_attn_linear(
 
     // Update the recurrent states
     ggml_build_forward_expand(gf,
-                              ggml_cpy(ctx0, new_state,
-                                       ggml_view_1d(ctx0, ssm_states_all, hparams.n_embd_s() * n_seqs,
-                                                    kv_head * hparams.n_embd_s() * ggml_element_size(ssm_states_all))));
+            ggml_cpy(ctx0, new_state,
+                ggml_view_1d(ctx0, ssm_states_all, hparams.n_embd_s() * n_seqs,
+                    kv_head * hparams.n_embd_s() * ggml_element_size(ssm_states_all))));
 
     // z: [head_dim, n_heads, n_tokens, n_seqs] -> [n_heads * n_tokens * n_seqs, head_dim]
     ggml_tensor * z_2d = ggml_reshape_4d(ctx0, z, head_v_dim, num_v_heads, n_seq_tokens, n_seqs);