kv-cache : support attention rotation for heterogeneous iSWA (#21513)

* kv-cache : support attention rotation for heterogeneous iSWA

* cont : remove assert

src/llama-kv-cache.h

@@ -239,6 +239,11 @@ private:
     bool attn_rot_k = false;
     bool attn_rot_v = false;
 
+    // if all layers participating in the cache have constant head size, the value is stored here
+    // otherwise the value is -1
+    int32_t n_embd_head_k_all = 0;
+    int32_t n_embd_head_v_all = 0;
+
     // pre-computed hadamard martrices
     std::unordered_map<int64_t, std::vector<float>> attn_rot_hadamard;