common : move up common_init() and fix Windows UTF-8 logs (#21176)

The build info is now only for debug, so we avoid the duplicate
with `--version`.

The UTF-8 setup at the beginning is needed to avoid logging
garbage on Windows.

Signed-off-by: Adrien Gallouët <angt@huggingface.co>

common/common.cpp

@@ -359,6 +359,11 @@ bool parse_cpu_mask(const std::string & mask, bool (&boolmask)[GGML_MAX_N_THREAD
 }
 
 void common_init() {
+#if defined(_WIN32)
+    SetConsoleOutputCP(CP_UTF8);
+    SetConsoleCP(CP_UTF8);
+#endif
+
     llama_log_set(common_log_default_callback, NULL);
 
 #ifdef NDEBUG
@@ -367,7 +372,7 @@ void common_init() {
     const char * build_type = " (debug)";
 #endif
 
-    LOG_INF("build: %d (%s) with %s for %s%s\n", LLAMA_BUILD_NUMBER, LLAMA_COMMIT, LLAMA_COMPILER, LLAMA_BUILD_TARGET, build_type);
+    LOG_DBG("build: %d (%s) with %s for %s%s\n", LLAMA_BUILD_NUMBER, LLAMA_COMMIT, LLAMA_COMPILER, LLAMA_BUILD_TARGET, build_type);
 }
 
 std::string common_params_get_system_info(const common_params & params) {

examples/batched/batched.cpp

@@ -24,12 +24,12 @@ int main(int argc, char ** argv) {
     params.prompt = "Hello my name is";
     params.n_predict = 32;
 
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_BATCHED, print_usage)) {
         return 1;
     }
 
-    common_init();
-
     // number of parallel batches
     int n_parallel = params.n_parallel;
 

examples/debug/debug.cpp

@@ -213,12 +213,12 @@ static bool run(llama_context * ctx, const common_params & params) {
 int main(int argc, char ** argv) {
     common_params params;
 
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_DEBUG, print_usage)) {
         return 1;
     }
 
-    common_init();
-
     llama_backend_init();
     llama_numa_init(params.numa);
 

examples/diffusion/diffusion-cli.cpp

@@ -545,11 +545,12 @@ int main(int argc, char ** argv) {
 
     common_params params;
 
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_DIFFUSION)) {
         return 1;
     }
 
-    common_init();
     llama_backend_init();
 
     llama_model_params model_params = llama_model_default_params();

examples/embedding/embedding.cpp

@@ -99,12 +99,12 @@ int main(int argc, char ** argv) {
 
     common_params params;
 
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_EMBEDDING)) {
         return 1;
     }
 
-    common_init();
-
     params.embedding = true;
 
     // get max number of sequences per batch

examples/eval-callback/eval-callback.cpp

@@ -37,12 +37,12 @@ int main(int argc, char ** argv) {
 
     common_params params;
 
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_COMMON)) {
         return 1;
     }
 
-    common_init();
-
     llama_backend_init();
     llama_numa_init(params.numa);
 

examples/idle/idle.cpp

@@ -19,12 +19,12 @@ static void print_usage(int /*argc*/, char ** argv) {
 int main(int argc, char ** argv) {
     common_params params;
 
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_COMMON, print_usage)) {
         return 1;
     }
 
-    common_init();
-
     // init LLM
 
     llama_backend_init();

examples/lookahead/lookahead.cpp

@@ -43,12 +43,12 @@ int main(int argc, char ** argv) {
 
     common_params params;
 
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_COMMON)) {
         return 1;
     }
 
-    common_init();
-
     const int W = 15; // lookahead window
     const int N = 5;  // n-gram size
     const int G = 15; // max verification n-grams

examples/lookup/lookup-create.cpp

@@ -12,6 +12,8 @@ int main(int argc, char ** argv){
 
     common_params params;
 
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_LOOKUP)) {
         return 1;
     }

examples/lookup/lookup-stats.cpp

@@ -18,12 +18,12 @@ int main(int argc, char ** argv){
 
     common_params params;
 
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_LOOKUP)) {
         return 1;
     }
 
-    common_init();
-
     const int n_draft = params.speculative.n_max;
 
     // init llama.cpp

examples/lookup/lookup.cpp

@@ -18,12 +18,12 @@ int main(int argc, char ** argv){
 
     common_params params;
 
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_LOOKUP)) {
         return 1;
     }
 
-    common_init();
-
     // max. number of additional tokens to draft if match is found
     const int n_draft = params.speculative.n_max;
 

examples/parallel/parallel.cpp

@@ -163,12 +163,12 @@ int main(int argc, char ** argv) {
     params.n_predict = 128;
     params.n_junk = 1;
 
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_PARALLEL)) {
         return 1;
     }
 
-    common_init();
-
     // number of simultaneous "clients" to simulate
     const int32_t n_clients = params.n_parallel;
 

examples/passkey/passkey.cpp

@@ -25,12 +25,12 @@ int main(int argc, char ** argv) {
     params.n_keep = 32;
     params.i_pos  = -1;
 
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_PASSKEY, print_usage)) {
         return 1;
     }
 
-    common_init();
-
     int n_junk = params.n_junk;
     int n_keep = params.n_keep;
     int n_grp  = params.grp_attn_n;

examples/retrieval/retrieval.cpp

@@ -117,12 +117,12 @@ int main(int argc, char ** argv) {
 
     common_params params;
 
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_RETRIEVAL, print_usage)) {
         return 1;
     }
 
-    common_init();
-
     // For BERT models, batch size must be equal to ubatch size
     params.n_ubatch = params.n_batch;
     params.embedding = true;

examples/save-load-state/save-load-state.cpp

@@ -17,6 +17,8 @@ int main(int argc, char ** argv) {
 
     const std::string_view state_file = "dump_state.bin";
 
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_COMMON)) {
         return 1;
     }
@@ -27,8 +29,6 @@ int main(int argc, char ** argv) {
         params.kv_unified = true;
     }
 
-    common_init();
-
     if (params.n_predict < 0) {
         params.n_predict = 16;
     }

examples/speculative-simple/speculative-simple.cpp

@@ -16,6 +16,8 @@ int main(int argc, char ** argv) {
 
     common_params params;
 
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_SPECULATIVE)) {
         return 1;
     }
@@ -25,8 +27,6 @@ int main(int argc, char ** argv) {
         return 1;
     }
 
-    common_init();
-
     if (params.speculative.mparams_dft.path.empty()) {
         LOG_ERR("%s: --model-draft is required\n", __func__);
         return 1;

examples/speculative/speculative.cpp

@@ -38,6 +38,8 @@ int main(int argc, char ** argv) {
     // needed to get candidate probs even for temp <= 0.0
     params.sampling.n_probs = 128;
 
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_SPECULATIVE)) {
         return 1;
     }
@@ -47,8 +49,6 @@ int main(int argc, char ** argv) {
         return 1;
     }
 
-    common_init();
-
     if (params.speculative.mparams_dft.path.empty()) {
         LOG_ERR("%s: --model-draft is required\n", __func__);
         return 1;

examples/training/finetune.cpp

@@ -20,6 +20,8 @@ int main(int argc, char ** argv) {
     common_params params;
     params.escape = false;
 
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_FINETUNE)) {
         return 1;
     }
@@ -38,7 +40,6 @@ int main(int argc, char ** argv) {
         params.cache_type_v = GGML_TYPE_F32;
     }
 
-    common_init();
     llama_backend_init();
     llama_numa_init(params.numa);
     // load the model and apply lora adapter, if any

tests/export-graph-ops.cpp

@@ -118,12 +118,12 @@ int main(int argc, char ** argv) {
     common_params params;
     params.out_file = "tests.txt";
 
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_EXPORT_GRAPH_OPS)) {
         return 1;
     }
 
-    common_init();
-
     // Load CPU-only
     ggml_backend_dev_t cpu_device = ggml_backend_dev_by_type(GGML_BACKEND_DEVICE_TYPE_CPU);
     params.devices = { cpu_device, nullptr };

tests/test-state-restore-fragmented.cpp

@@ -22,12 +22,12 @@ int main(int argc, char ** argv) {
     params.n_parallel = 3;
     params.n_ctx = 256;
 
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_COMMON)) {
         return 1;
     }
 
-    common_init();
-
     // init
     common_init_result_ptr llama_init = common_init_from_params(params);
 

tests/test-thread-safety.cpp

@@ -16,12 +16,12 @@
 int main(int argc, char ** argv) {
     common_params params;
 
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_COMMON)) {
         return 1;
     }
 
-    common_init();
-
     llama_backend_init();
     llama_numa_init(params.numa);
 

tools/batched-bench/batched-bench.cpp

@@ -20,12 +20,12 @@ int main(int argc, char ** argv) {
 
     common_params params;
 
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_BENCH, print_usage)) {
         return 1;
     }
 
-    common_init();
-
     int is_pp_shared   = params.is_pp_shared;
     int is_tg_separate = params.is_tg_separate;
 

tools/cli/cli.cpp

@@ -347,6 +347,8 @@ int main(int argc, char ** argv) {
 
     params.verbosity = LOG_LEVEL_ERROR; // by default, less verbose logs
 
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_CLI)) {
         return 1;
     }
@@ -357,8 +359,6 @@ int main(int argc, char ** argv) {
         console::error("please use llama-completion instead\n");
     }
 
-    common_init();
-
     // struct that contains llama context and inference
     cli_context ctx_cli(params);
 

tools/completion/completion.cpp

@@ -90,12 +90,12 @@ int main(int argc, char ** argv) {
     common_params params;
     g_params = &params;
 
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_COMPLETION, print_usage)) {
         return 1;
     }
 
-    common_init();
-
     auto & sparams = params.sampling;
 
     // save choice to use color for later

tools/cvector-generator/cvector-generator.cpp

@@ -400,6 +400,8 @@ int main(int argc, char ** argv) {
 
     params.out_file = "control_vector.gguf";
 
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_CVECTOR_GENERATOR, print_usage)) {
         return 1;
     }

tools/export-lora/export-lora.cpp

@@ -418,6 +418,8 @@ int main(int argc, char ** argv) {
 
     params.out_file = "ggml-lora-merged-f16.gguf";
 
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_EXPORT_LORA, print_usage)) {
         return 1;
     }

tools/fit-params/fit-params.cpp

@@ -17,11 +17,12 @@ using namespace std::chrono_literals;
 int main(int argc, char ** argv) {
     common_params params;
 
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_COMMON)) {
         return 1;
     }
 
-    common_init();
     llama_backend_init();
     llama_numa_init(params.numa);
     auto mparams = common_model_params_to_llama(params);

tools/imatrix/imatrix.cpp

@@ -1212,6 +1212,8 @@ int main(int argc, char ** argv) {
     params.n_ctx = 512;
     params.escape = false;
 
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_IMATRIX, print_usage)) {
         return 1;
     }
@@ -1223,8 +1225,6 @@ int main(int argc, char ** argv) {
         return 0;
     }
 
-    common_init();
-
     const int32_t n_ctx = params.n_ctx;
 
     if (n_ctx <= 0) {

tools/mtmd/debug/mtmd-debug.cpp

@@ -54,11 +54,12 @@ int main(int argc, char ** argv) {
 
     common_params params;
 
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_MTMD, show_additional_info)) {
         return 1;
     }
 
-    common_init();
     mtmd_helper_log_set(common_log_default_callback, nullptr);
 
     if (params.mmproj.path.empty()) {

tools/mtmd/mtmd-cli.cpp

@@ -281,11 +281,12 @@ int main(int argc, char ** argv) {
 
     common_params params;
 
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_MTMD, show_additional_info)) {
         return 1;
     }
 
-    common_init();
     mtmd_helper_log_set(common_log_default_callback, nullptr);
 
     if (params.mmproj.path.empty()) {

tools/perplexity/perplexity.cpp

@@ -2012,12 +2012,12 @@ int main(int argc, char ** argv) {
     params.n_ctx = 512;
     params.escape = false;
 
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_PERPLEXITY)) {
         return 1;
     }
 
-    common_init();
-
     const int32_t n_ctx = params.n_ctx;
 
     if (n_ctx <= 0) {

tools/results/results.cpp

@@ -58,6 +58,9 @@ static std::vector<float> get_logits(
 int main(int argc, char ** argv) {
     common_params params;
     params.escape = false;
+
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_RESULTS)) {
         return 1;
     }
@@ -65,7 +68,6 @@ int main(int argc, char ** argv) {
         LOG_ERR("%s: an output file must be specified", __func__);
         return 1;
     }
-    common_init();
     llama_backend_init();
     llama_numa_init(params.numa);
     common_init_result_ptr llama_init = common_init_from_params(params);

tools/server/server.cpp

@@ -75,6 +75,8 @@ int main(int argc, char ** argv) {
     // own arguments required by this example
     common_params params;
 
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_SERVER)) {
         return 1;
     }
@@ -100,8 +102,6 @@ int main(int argc, char ** argv) {
         params.model_alias.insert(params.model.name);
     }
 
-    common_init();
-
     // struct that contains llama context and inference
     server_context ctx_server;
 

tools/tts/tts.cpp

@@ -551,6 +551,8 @@ int main(int argc, char ** argv) {
     params.sampling.top_k = 4;
     params.sampling.samplers = { COMMON_SAMPLER_TYPE_TOP_K, };
 
+    common_init();
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_TTS, print_usage)) {
         return 1;
     }
@@ -558,8 +560,6 @@ int main(int argc, char ** argv) {
     const int n_parallel = params.n_parallel;
     const int n_predict  = params.n_predict;
 
-    common_init();
-
     // init LLM
 
     llama_backend_init();