llama : add llama_supports_rpc API

Author: slaren

common/arg.cpp

@@ -1353,15 +1353,15 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
             params.image.emplace_back(value);
         }
     ).set_examples({LLAMA_EXAMPLE_LLAVA}));
-#ifdef GGML_USE_RPC
-    add_opt(llama_arg(
-        {"--rpc"}, "SERVERS",
-        "comma separated list of RPC servers",
-        [](gpt_params & params, const std::string & value) {
-            params.rpc_servers = value;
-        }
-    ).set_env("LLAMA_ARG_RPC"));
-#endif
+    if (llama_supports_rpc()) {
+        add_opt(llama_arg(
+            {"--rpc"}, "SERVERS",
+            "comma separated list of RPC servers",
+            [](gpt_params & params, const std::string & value) {
+                params.rpc_servers = value;
+            }
+        ).set_env("LLAMA_ARG_RPC"));
+    }
     add_opt(llama_arg(
         {"--mlock"},
         "force system to keep model in RAM rather than swapping or compressing",

examples/llama-bench/llama-bench.cpp

@@ -304,9 +304,9 @@ static void print_usage(int /* argc */, char ** argv) {
     printf("  --cpu-strict <0|1>                        (default: %s)\n", join(cmd_params_defaults.cpu_strict, ",").c_str());
     printf("  --poll <0...100>                          (default: %s)\n", join(cmd_params_defaults.poll, ",").c_str());
     printf("  -ngl, --n-gpu-layers <n>                  (default: %s)\n", join(cmd_params_defaults.n_gpu_layers, ",").c_str());
-#ifdef GGML_USE_RPC
-    printf("  -rpc, --rpc <rpc_servers>                 (default: %s)\n", join(cmd_params_defaults.rpc_servers, ",").c_str());
-#endif
+    if (llama_supports_rpc()) {
+        printf("  -rpc, --rpc <rpc_servers>                 (default: %s)\n", join(cmd_params_defaults.rpc_servers, ",").c_str());
+    }
     printf("  -sm, --split-mode <none|layer|row>        (default: %s)\n", join(transform_to_str(cmd_params_defaults.split_mode, split_mode_str), ",").c_str());
     printf("  -mg, --main-gpu <i>                       (default: %s)\n", join(cmd_params_defaults.main_gpu, ",").c_str());
     printf("  -nkvo, --no-kv-offload <0|1>              (default: %s)\n", join(cmd_params_defaults.no_kv_offload, ",").c_str());
@@ -497,14 +497,12 @@ static cmd_params parse_cmd_params(int argc, char ** argv) {
             }
             auto p = string_split<int>(argv[i], split_delim);
             params.n_gpu_layers.insert(params.n_gpu_layers.end(), p.begin(), p.end());
-#ifdef GGML_USE_RPC
-        } else if (arg == "-rpc" || arg == "--rpc") {
+        } else if (llama_supports_rpc() && (arg == "-rpc" || arg == "--rpc")) {
             if (++i >= argc) {
                 invalid_param = true;
                 break;
             }
             params.rpc_servers.push_back(argv[i]);
-#endif
         } else if (arg == "-sm" || arg == "--split-mode") {
             if (++i >= argc) {
                 invalid_param = true;

include/llama.h

@@ -433,6 +433,7 @@ extern "C" {
     LLAMA_API bool llama_supports_mmap       (void);
     LLAMA_API bool llama_supports_mlock      (void);
     LLAMA_API bool llama_supports_gpu_offload(void);
+    LLAMA_API bool llama_supports_rpc        (void);
 
     LLAMA_API uint32_t llama_n_ctx      (const struct llama_context * ctx);
     LLAMA_API uint32_t llama_n_batch    (const struct llama_context * ctx);

src/llama.cpp

@@ -18996,10 +18996,14 @@ bool llama_supports_gpu_offload(void) {
 #else
     return ggml_backend_dev_by_type(GGML_BACKEND_DEVICE_TYPE_GPU) != nullptr ||
            ggml_backend_dev_by_type(GGML_BACKEND_DEVICE_TYPE_GPU_FULL) != nullptr ||
-           ggml_backend_reg_by_name("RPC") != nullptr;
+           llama_supports_rpc();
 #endif
 }
 
+bool llama_supports_rpc(void) {
+    return ggml_backend_reg_by_name("RPC") != nullptr;
+}
+
 void llama_backend_init(void) {
     ggml_time_init();