Document the new max GPU layers default in help

This is a key change, just letting users know.

Signed-off-by: Eric Curtin <ericcurtin17@gmail.com>

Author: ericcurtin

common/arg.cpp

@@ -2466,7 +2466,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
     ).set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_N_CPU_MOE_DRAFT"));
     add_opt(common_arg(
         {"-ngl", "--gpu-layers", "--n-gpu-layers"}, "N",
-        "number of layers to store in VRAM",
+        string_format("number of layers to store in VRAM (default: %d, 999 = max. number of model layers to store in VRAM)", params.n_gpu_layers),
         [](common_params & params, int value) {
             params.n_gpu_layers = value;
             if (!llama_supports_gpu_offload()) {