Fix wrong shape for K norm

Author: pwilkin

src/models/llm_build_qwen3next.cpp

@@ -173,15 +173,14 @@ struct ggml_tensor * llm_build_qwen3next::build_qwen3next_attention_layer(ggml_t
     cb(Vcur, "Vcur", il);
 
     // Apply K normalization
+    Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
     Kcur = build_q3n_norm(Kcur, model.layers[il].attn_k_norm, il);
     cb(Kcur, "Kcur_normed", il);
 
     // Reshape gate to [n_embd, n_tokens] for the sigmoid gating (flatten the heads)
     gate = ggml_cont_2d(ctx0, gate, n_embd_head * n_head, n_tokens);
     cb(gate, "gate_reshaped", il);
 
-    Qcur = ggml_cont_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens);
-    Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
     Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
 
     // Apply RoPE
@@ -204,7 +203,6 @@ struct ggml_tensor * llm_build_qwen3next::build_qwen3next_attention_layer(ggml_t
     struct ggml_tensor * gate_sigmoid = ggml_sigmoid(ctx0, gate);
     cb(gate_sigmoid, "gate_sigmoid", il);
 
-    // Apply gating directly using the original gate tensor
     cur = ggml_mul(ctx0, cur, gate_sigmoid);
     cb(cur, "attn_gated", il);