llama : disable per-tensor info prints on model load (#4562)

ggml-cuda.cu

@@ -9091,7 +9091,7 @@ void ggml_cuda_transform_tensor(void * data, struct ggml_tensor * tensor) {
 }
 
 void ggml_cuda_free_data(struct ggml_tensor * tensor) {
-    if (!tensor || (tensor->backend != GGML_BACKEND_GPU && tensor->backend != GGML_BACKEND_GPU_SPLIT) ) {
+    if (!tensor || !tensor->extra || (tensor->backend != GGML_BACKEND_GPU && tensor->backend != GGML_BACKEND_GPU_SPLIT) ) {
         return;
     }
 

llama.cpp

@@ -2083,7 +2083,7 @@ struct llama_model_loader {
                     type_max   = meta->type;
                 }
 
-                LLAMA_LOG_INFO("%s: - tensor %4d: %32s %-8s [ %s ]\n", __func__, i, name, ggml_type_name(meta->type), llama_format_tensor_shape(meta).c_str());
+                // LLAMA_LOG_INFO("%s: - tensor %4d: %32s %-8s [ %s ]\n", __func__, i, name, ggml_type_name(meta->type), llama_format_tensor_shape(meta).c_str());
             }
 
             switch (type_max) {