model : make Gemma 4 shared-KV tail attn_k tensors optional on load (#21739)

diff --git a/src/llama-model.cpp b/src/llama-model.cpp
index 82af6b6bee33de0f1c0d02e2d3ba9ebff826fee6..f057fd31c351516fb3f68685aa83c5dac1bada69 100644 (file)
--- a/src/llama-model.cpp
+++ b/src/llama-model.cpp
@@ -4623,17 +4623,18 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
                         const int64_t n_embd_head = hparams.n_embd_head_k(i);
                         const int64_t n_embd_k    = hparams.n_embd_k_gqa(i);
                         const int64_t n_embd_v    = hparams.n_embd_v_gqa(i);
+                        const int     kv_flags    = hparams.has_kv(i) ? 0 : TENSOR_NOT_REQUIRED;
 
                         layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
 
                         // note: use_alternative_attention (v_proj is optional, if it's not present, use k_proj)
                         layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q,   "weight", i), {n_embd, n_embd_head * n_head}, 0);
-                        layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "weight", i), {n_embd, n_embd_k}, 0);
+                        layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K,   "weight", i), {n_embd, n_embd_k}, kv_flags);
                         layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V,   "weight", i), {n_embd, n_embd_v}, TENSOR_NOT_REQUIRED);
                         layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd_head * n_head, n_embd}, 0);
 
                         layer.attn_q_norm    = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM,    "weight", i), {n_embd_head}, 0);
-                        layer.attn_k_norm    = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM,    "weight", i), {n_embd_head}, 0);
+                        layer.attn_k_norm    = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM,    "weight", i), {n_embd_head}, kv_flags);
                         layer.attn_post_norm = create_tensor(tn(LLM_TENSOR_ATTN_POST_NORM, "weight", i), {n_embd}, 0);
 
                         layer.out_scale = create_tensor(tn(LLM_TENSOR_LAYER_OUT_SCALE, "weight", i), {1u}, TENSOR_NOT_REQUIRED);