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>

examples/batched/batched.cpp

@@ -5,6 +5,7 @@
 #include "sampling.h"
 
 #include <algorithm>
+#include <clocale>
 #include <cstdio>
 #include <string>
 #include <vector>
@@ -16,6 +17,8 @@ static void print_usage(int, char ** argv) {
 }
 
 int main(int argc, char ** argv) {
+    std::setlocale(LC_NUMERIC, "C");
+
     common_params params;
 
     params.prompt = "Hello my name is";

examples/convert-llama2c-to-ggml/convert-llama2c-to-ggml.cpp

@@ -5,14 +5,16 @@
 #include "common.h"
 #include "log.h"
 
+#include <algorithm>
+#include <cassert>
+#include <cinttypes>
+#include <climits>
+#include <clocale>
+#include <cstdarg>
+#include <cstring>
+#include <ctime>
 #include <unordered_map>
 #include <vector>
-#include <cassert>
-#include <climits>
-#include <cstring>
-#include <cstdarg>
-#include <cinttypes>
-#include <ctime>
 #include <random>
 #include <stdexcept>
 #include <sstream>
@@ -874,6 +876,8 @@ static std::string basename(const std::string &path) {
 }
 
 int main(int argc, char ** argv) {
+    std::setlocale(LC_NUMERIC, "C");
+
     common_init();
 
     struct train_params params = get_default_train_params();

examples/deprecation-warning/deprecation-warning.cpp

@@ -1,11 +1,14 @@
 // Warns users that this filename was deprecated, and provides a link for more information.
 
+#include <clocale>
 #include <cstdio>
 #include <string>
 #include <unordered_map>
 
 // Main
 int main(int argc, char** argv) {
+    std::setlocale(LC_NUMERIC, "C");
+
     std::string filename = "main";
     if (argc >= 1) {
         filename = argv[0];

examples/diffusion/diffusion-cli.cpp

@@ -7,6 +7,7 @@
 #include <limits.h>
 
 #include <algorithm>
+#include <clocale>
 #include <cmath>
 #include <cstring>
 #include <limits>
@@ -538,6 +539,8 @@ static std::string format_input_text(const std::string & prompt, const std::stri
 }
 
 int main(int argc, char ** argv) {
+    std::setlocale(LC_NUMERIC, "C");
+
     ggml_time_init();
 
     common_params params;

examples/embedding/embedding.cpp

@@ -3,6 +3,7 @@
 #include "log.h"
 #include "llama.h"
 
+#include <clocale>
 #include <ctime>
 #include <algorithm>
 
@@ -94,6 +95,8 @@ static void print_raw_embeddings(const float * emb,
 }
 
 int main(int argc, char ** argv) {
+    std::setlocale(LC_NUMERIC, "C");
+
     common_params params;
 
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_EMBEDDING)) {

examples/eval-callback/eval-callback.cpp

@@ -4,6 +4,8 @@
 #include "log.h"
 #include "llama.h"
 #include "llama-cpp.h"
+
+#include <clocale>
 #include <string>
 #include <vector>
 
@@ -29,6 +31,8 @@ static bool run(llama_context * ctx, const common_params & params) {
 }
 
 int main(int argc, char ** argv) {
+    std::setlocale(LC_NUMERIC, "C");
+
     base_callback_data cb_data;
 
     common_params params;

examples/gen-docs/gen-docs.cpp

@@ -1,6 +1,7 @@
 #include "arg.h"
 #include "common.h"
 
+#include <clocale>
 #include <fstream>
 #include <sstream>
 #include <string>
@@ -100,6 +101,8 @@ static void write_help(std::ostringstream & ss, const md_file & md) {
 }
 
 int main(int, char **) {
+    std::setlocale(LC_NUMERIC, "C");
+
     for (const auto & md : md_files) {
         std::ifstream infile(md.fname);
         if (!infile.is_open()) {

examples/gguf-hash/gguf-hash.cpp

@@ -1,13 +1,14 @@
 #include "ggml.h"
 #include "gguf.h"
 
-#include <cstdlib>   /* abort() */
+#include <algorithm>
+#include <clocale>
 #include <cstddef>
 #include <cstdio>
-#include <string>
+#include <cstdlib>   /* abort() */
-#include <stdexcept>
-#include <algorithm>
 #include <cstring>
+#include <stdexcept>
+#include <string>
 
 #include <sstream>
 #include <fstream>
@@ -626,6 +627,8 @@ static hash_exit_code_t gguf_hash(const hash_params & hash_params) {
 }
 
 int main(int argc, const char ** argv) {
+    std::setlocale(LC_NUMERIC, "C");
+
     hash_params params;
     manifest_check_params manifest_check;
     hash_params_parse(argc, argv, params);

examples/gguf/gguf.cpp

@@ -1,6 +1,7 @@
 #include "ggml.h"
 #include "gguf.h"
 
+#include <clocale>
 #include <cstdio>
 #include <string>
 #include <sstream>
@@ -240,6 +241,8 @@ static bool gguf_ex_read_1(const std::string & fname, bool check_data) {
 }
 
 int main(int argc, char ** argv) {
+    std::setlocale(LC_NUMERIC, "C");
+
     if (argc < 3) {
         printf("usage: %s data.gguf r|w [n]\n", argv[0]);
         printf("r: read data.gguf file\n");

examples/lookahead/lookahead.cpp

@@ -4,10 +4,11 @@
 #include "log.h"
 #include "llama.h"
 
+#include <algorithm>
+#include <clocale>
 #include <cstdio>
 #include <string>
 #include <vector>
-#include <algorithm>
 
 struct ngram_data {
     bool active = false;
@@ -38,6 +39,8 @@ struct ngram_container {
 };
 
 int main(int argc, char ** argv) {
+    std::setlocale(LC_NUMERIC, "C");
+
     common_params params;
 
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_COMMON)) {

examples/lookup/lookup-create.cpp

@@ -3,10 +3,13 @@
 #include "ngram-cache.h"
 #include "llama.h"
 
+#include <clocale>
 #include <string>
 #include <vector>
 
 int main(int argc, char ** argv){
+    std::setlocale(LC_NUMERIC, "C");
+
     common_params params;
 
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_LOOKUP)) {

examples/lookup/lookup-merge.cpp

@@ -3,6 +3,7 @@
 #include "common.h"
 #include "ngram-cache.h"
 
+#include <clocale>
 #include <cstdint>
 #include <cstdio>
 #include <fstream>
@@ -17,6 +18,8 @@ static void print_usage(char* argv0) {
 }
 
 int main(int argc, char ** argv){
+    std::setlocale(LC_NUMERIC, "C");
+
     if (argc < 3) {
         print_usage(argv[0]);
         exit(1);

examples/lookup/lookup-stats.cpp

@@ -5,14 +5,17 @@
 #include "llama.h"
 #include "ggml.h"
 
+#include <cinttypes>
+#include <clocale>
 #include <cstdint>
 #include <cstdio>
-#include <cinttypes>
 #include <fstream>
 #include <string>
 #include <vector>
 
 int main(int argc, char ** argv){
+    std::setlocale(LC_NUMERIC, "C");
+
     common_params params;
 
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_LOOKUP)) {

examples/lookup/lookup.cpp

@@ -6,6 +6,7 @@
 #include "log.h"
 #include "llama.h"
 
+#include <clocale>
 #include <cstdint>
 #include <cstdio>
 #include <fstream>
@@ -13,6 +14,8 @@
 #include <vector>
 
 int main(int argc, char ** argv){
+    std::setlocale(LC_NUMERIC, "C");
+
     common_params params;
 
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_LOOKUP)) {

examples/parallel/parallel.cpp

@@ -7,12 +7,13 @@
 #include "log.h"
 #include "llama.h"
 
+#include <algorithm>
+#include <clocale>
 #include <cmath>
 #include <cstdio>
 #include <string>
 #include <vector>
 #include <ctime>
-#include <algorithm>
 
 // trim whitespace from the beginning and end of a string
 static std::string trim(const std::string & str) {
@@ -153,6 +154,8 @@ static std::vector<std::string> split_string(const std::string& input, char deli
 }
 
 int main(int argc, char ** argv) {
+    std::setlocale(LC_NUMERIC, "C");
+
     srand(1234);
 
     common_params params;

examples/passkey/passkey.cpp

@@ -3,6 +3,7 @@
 #include "log.h"
 #include "llama.h"
 
+#include <clocale>
 #include <cmath>
 #include <cstdio>
 #include <string>
@@ -16,6 +17,8 @@ static void print_usage(int, char ** argv) {
 }
 
 int main(int argc, char ** argv) {
+    std::setlocale(LC_NUMERIC, "C");
+
     common_params params;
 
     params.n_junk = 250;

examples/retrieval/retrieval.cpp

@@ -4,6 +4,7 @@
 #include "llama.h"
 
 #include <algorithm>
+#include <clocale>
 #include <fstream>
 #include <iostream> // TODO: remove me
 
@@ -112,6 +113,8 @@ static void batch_process(llama_context * ctx, llama_batch & batch, float * outp
 }
 
 int main(int argc, char ** argv) {
+    std::setlocale(LC_NUMERIC, "C");
+
     common_params params;
 
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_RETRIEVAL, print_usage)) {

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

@@ -2,11 +2,14 @@
 #include "common.h"
 #include "llama.h"
 
+#include <clocale>
 #include <vector>
 #include <cstdio>
 
 
 int main(int argc, char ** argv) {
+    std::setlocale(LC_NUMERIC, "C");
+
     common_params params;
 
     params.prompt = "The quick brown fox";

examples/simple-chat/simple-chat.cpp

@@ -1,4 +1,5 @@
 #include "llama.h"
+#include <clocale>
 #include <cstdio>
 #include <cstring>
 #include <iostream>
@@ -12,6 +13,8 @@ static void print_usage(int, char ** argv) {
 }
 
 int main(int argc, char ** argv) {
+    std::setlocale(LC_NUMERIC, "C");
+
     std::string model_path;
     int ngl = 99;
     int n_ctx = 2048;

examples/simple/simple.cpp

@@ -1,4 +1,5 @@
 #include "llama.h"
+#include <clocale>
 #include <cstdio>
 #include <cstring>
 #include <string>
@@ -11,6 +12,8 @@ static void print_usage(int, char ** argv) {
 }
 
 int main(int argc, char ** argv) {
+    std::setlocale(LC_NUMERIC, "C");
+
     // path to the model gguf file
     std::string model_path;
     // prompt to generate text from

examples/speculative-simple/speculative-simple.cpp

@@ -5,12 +5,15 @@
 #include "log.h"
 #include "llama.h"
 
+#include <clocale>
 #include <cstdio>
 #include <cstring>
 #include <string>
 #include <vector>
 
 int main(int argc, char ** argv) {
+    std::setlocale(LC_NUMERIC, "C");
+
     common_params params;
 
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_SPECULATIVE)) {

examples/speculative/speculative.cpp

@@ -5,6 +5,7 @@
 #include "llama.h"
 
 #include <algorithm>
+#include <clocale>
 #include <cstdio>
 #include <cstring>
 #include <random>
@@ -30,6 +31,8 @@ struct seq_draft {
 };
 
 int main(int argc, char ** argv) {
+    std::setlocale(LC_NUMERIC, "C");
+
     common_params params;
 
     // needed to get candidate probs even for temp <= 0.0

examples/sycl/ls-sycl-device.cpp

@@ -6,8 +6,10 @@
 
 
 #include "ggml-sycl.h"
+#include <clocale>
 
 int main() {
+    std::setlocale(LC_NUMERIC, "C");
     ggml_backend_sycl_print_sycl_devices();
     return 0;
 }

examples/training/finetune.cpp

@@ -3,6 +3,7 @@
 #include "log.h"
 #include "llama.h"
 
+#include <clocale>
 #include <cmath>
 #include <cstdio>
 #include <cstring>
@@ -14,6 +15,8 @@
 #endif
 
 int main(int argc, char ** argv) {
+    std::setlocale(LC_NUMERIC, "C");
+
     common_params params;
     params.escape = false;
 

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";