ggml : allow ggml_init with 0 size

Author: ggerganov

ggml.c

@@ -4579,6 +4579,11 @@ struct ggml_context * ggml_init(struct ggml_init_params params) {
         return NULL;
     }
 
+    // allow to call ggml_init with 0 size
+    if (params.mem_size == 0) {
+        params.mem_size = GGML_MEM_ALIGN;
+    }
+    
     const size_t mem_size = params.mem_buffer ? params.mem_size : GGML_PAD(params.mem_size, GGML_MEM_ALIGN);
 
     *ctx = (struct ggml_context) {

llama.cpp

@@ -5406,7 +5406,7 @@ void llama_backend_init(bool numa) {
 
     // needed to initialize f16 tables
     {
-        struct ggml_init_params params = { 1, NULL, false };
+        struct ggml_init_params params = { 0, NULL, false };
         struct ggml_context * ctx = ggml_init(params);
         ggml_free(ctx);
     }