Fix locale-dependent float printing in GGUF metadata (#17331)

* Set C locale for consistent float formatting across all binaries.

* Add C locale setting to all tools binaries

Add std::setlocale(LC_NUMERIC, "C") to all 16 binaries in the tools/
directory to ensure consistent floating-point formatting.

* Apply suggestion from @JohannesGaessler

---------

Co-authored-by: Johannes Gäßler <johannesg@5d6.de>

tools/batched-bench/batched-bench.cpp

@@ -4,6 +4,7 @@
 #include "llama.h"
 
 #include <algorithm>
+#include <clocale>
 #include <cstdio>
 #include <string>
 #include <vector>
@@ -15,6 +16,8 @@ static void print_usage(int, char ** argv) {
 }
 
 int main(int argc, char ** argv) {
+    std::setlocale(LC_NUMERIC, "C");
+
     common_params params;
 
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_BENCH, print_usage)) {

tools/completion/completion.cpp

@@ -6,6 +6,7 @@
 #include "llama.h"
 #include "chat.h"
 
+#include <clocale>
 #include <cstdio>
 #include <cstring>
 #include <ctime>
@@ -84,6 +85,8 @@ static void sigint_handler(int signo) {
 #endif
 
 int main(int argc, char ** argv) {
+    std::setlocale(LC_NUMERIC, "C");
+
     common_params params;
     g_params = &params;
 

tools/cvector-generator/cvector-generator.cpp

@@ -7,6 +7,8 @@
 #include "pca.hpp"
 #include "mean.hpp"
 
+#include <clocale>
+
 #ifdef GGML_USE_CUDA
 #include "ggml-cuda.h"
 #endif
@@ -392,6 +394,8 @@ static int prepare_entries(common_params & params, train_context & ctx_train) {
 }
 
 int main(int argc, char ** argv) {
+    std::setlocale(LC_NUMERIC, "C");
+
     common_params params;
 
     params.out_file = "control_vector.gguf";

tools/export-lora/export-lora.cpp

@@ -5,6 +5,7 @@
 #include "arg.h"
 #include "common.h"
 
+#include <clocale>
 #include <map>
 #include <vector>
 #include <string>
@@ -411,6 +412,8 @@ static void print_usage(int, char ** argv) {
 }
 
 int main(int argc, char ** argv) {
+    std::setlocale(LC_NUMERIC, "C");
+
     common_params params;
 
     params.out_file = "ggml-lora-merged-f16.gguf";

tools/gguf-split/gguf-split.cpp

@@ -6,6 +6,7 @@
 #include <algorithm>
 #include <cinttypes>
 #include <climits>
+#include <clocale>
 #include <cstdio>
 #include <cstdlib>
 #include <stdexcept>
@@ -567,6 +568,8 @@ static void gguf_merge(const split_params & split_params) {
 }
 
 int main(int argc, const char ** argv) {
+    std::setlocale(LC_NUMERIC, "C");
+
     split_params params;
     split_params_parse(argc, argv, params);
 

tools/imatrix/imatrix.cpp

@@ -6,6 +6,7 @@
 
 #include <algorithm>
 #include <chrono>
+#include <clocale>
 #include <cmath>
 #include <cstdio>
 #include <cstring>
@@ -1191,6 +1192,8 @@ static bool show_statistics(const common_params & params) {
 }
 
 int main(int argc, char ** argv) {
+    std::setlocale(LC_NUMERIC, "C");
+
     common_params params;
 
     params.out_file = "imatrix.gguf";

tools/llama-bench/llama-bench.cpp

@@ -2034,8 +2034,9 @@ static std::unique_ptr<printer> create_printer(output_formats format) {
 }
 
 int main(int argc, char ** argv) {
+    std::setlocale(LC_NUMERIC, "C");
     // try to set locale for unicode characters in markdown
-    setlocale(LC_CTYPE, ".UTF-8");
+    std::setlocale(LC_CTYPE, ".UTF-8");
 
 #if !defined(NDEBUG)
     fprintf(stderr, "warning: asserts enabled, performance may be affected\n");

tools/mtmd/deprecation-warning.cpp

@@ -1,7 +1,10 @@
+#include <clocale>
 #include <cstdio>
 #include <string>
 
 int main(int argc, char** argv) {
+    std::setlocale(LC_NUMERIC, "C");
+
     std::string filename = "main";
     if (argc >= 1) {
         filename = argv[0];

tools/mtmd/mtmd-cli.cpp

@@ -274,6 +274,8 @@ static int eval_message(mtmd_cli_context & ctx, common_chat_msg & msg) {
 }
 
 int main(int argc, char ** argv) {
+    std::setlocale(LC_NUMERIC, "C");
+
     ggml_time_init();
 
     common_params params;

tools/perplexity/perplexity.cpp

@@ -3,10 +3,11 @@
 #include "log.h"
 #include "llama.h"
 
-#include <chrono>
 #include <algorithm>
 #include <array>
 #include <atomic>
+#include <chrono>
+#include <clocale>
 #include <cmath>
 #include <cstdio>
 #include <cstring>
@@ -2004,6 +2005,8 @@ static void kl_divergence(llama_context * ctx, const common_params & params) {
 }
 
 int main(int argc, char ** argv) {
+    std::setlocale(LC_NUMERIC, "C");
+
     common_params params;
 
     params.n_ctx = 512;

tools/quantize/quantize.cpp

@@ -2,6 +2,10 @@
 #include "llama.h"
 #include "gguf.h"
 
+#include <algorithm>
+#include <cctype>
+#include <clocale>
+#include <cmath>
 #include <cstdio>
 #include <cstring>
 #include <vector>
@@ -485,6 +489,8 @@ static bool parse_layer_prune(const char * data, std::vector<int> & prune_layers
 }
 
 int main(int argc, char ** argv) {
+    std::setlocale(LC_NUMERIC, "C");
+
     if (argc < 3) {
         usage(argv[0]);
     }

tools/rpc/rpc-server.cpp

@@ -10,12 +10,15 @@
 #  include <unistd.h>
 #  include <sys/stat.h>
 #endif
-#include <string>
-#include <stdio.h>
-#include <vector>
 #include <algorithm>
-#include <thread>
+#include <clocale>
+#include <codecvt>
+#include <filesystem>
 #include <regex>
+#include <stdio.h>
+#include <string>
+#include <thread>
+#include <vector>
 
 #if defined(__linux__)
 #include <sys/types.h>
@@ -285,6 +288,8 @@ static std::vector<ggml_backend_dev_t> get_devices(const rpc_server_params & par
 }
 
 int main(int argc, char * argv[]) {
+    std::setlocale(LC_NUMERIC, "C");
+
     ggml_backend_load_all();
 
     rpc_server_params params;

tools/server/server.cpp

@@ -8,6 +8,7 @@
 #include "log.h"
 
 #include <atomic>
+#include <clocale>
 #include <exception>
 #include <signal.h>
 #include <thread> // for std::thread::hardware_concurrency
@@ -67,6 +68,8 @@ static server_http_context::handler_t ex_wrapper(server_http_context::handler_t
 }
 
 int main(int argc, char ** argv) {
+    std::setlocale(LC_NUMERIC, "C");
+
     // own arguments required by this example
     common_params params;
 

tools/tokenize/tokenize.cpp

@@ -2,6 +2,7 @@
 //#include "log.h" // TODO: start using log.h
 #include "llama.h"
 
+#include <clocale>
 #include <cstdio>
 #include <cstring>
 #include <fstream>
@@ -184,6 +185,8 @@ static void write_utf8_cstr_to_stdout(const char * str, bool & invalid_utf8) {
 }
 
 int main(int raw_argc, char ** raw_argv) {
+    std::setlocale(LC_NUMERIC, "C");
+
     const std::vector<std::string> argv = ingest_args(raw_argc, raw_argv);
     const int argc = argv.size();
 

tools/tts/tts.cpp

@@ -10,6 +10,7 @@
 #include <nlohmann/json.hpp>
 
 #include <algorithm>
+#include <clocale>
 #include <cmath>
 #include <cstdio>
 #include <fstream>
@@ -536,6 +537,8 @@ static std::string audio_data_from_speaker(json speaker, const outetts_version t
 }
 
 int main(int argc, char ** argv) {
+    std::setlocale(LC_NUMERIC, "C");
+
     common_params params;
 
     params.out_file = "output.wav";