talk-llama : sync llama.cpp

examples/talk-llama/models/llama.cpp

@@ -124,6 +124,8 @@ llama_model_llama::graph<embed>::graph(const llama_model & model, const llm_grap
     ggml_tensor * inp_out_ids = build_inp_out_ids();
 
     for (int il = 0; il < n_layer; ++il) {
+        res->t_layer_inp[il] = inpL;
+
         ggml_tensor * inpSA = inpL;
 
         // norm