diff --git a/examples/batched/batched.cpp b/examples/batched/batched.cpp
index a8c19a6aba..d2b2e336e7 100644
--- a/examples/batched/batched.cpp
+++ b/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";
diff --git a/examples/convert-llama2c-to-ggml/convert-llama2c-to-ggml.cpp b/examples/convert-llama2c-to-ggml/convert-llama2c-to-ggml.cpp
index 767198aafa..702bc74bee 100644
--- a/examples/convert-llama2c-to-ggml/convert-llama2c-to-ggml.cpp
+++ b/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();
diff --git a/examples/deprecation-warning/deprecation-warning.cpp b/examples/deprecation-warning/deprecation-warning.cpp
index 11f5147328..0cde17f6e9 100644
--- a/examples/deprecation-warning/deprecation-warning.cpp
+++ b/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];
diff --git a/examples/diffusion/diffusion-cli.cpp b/examples/diffusion/diffusion-cli.cpp
index d50f754092..d38bfe7f82 100644
--- a/examples/diffusion/diffusion-cli.cpp
+++ b/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;
diff --git a/examples/embedding/embedding.cpp b/examples/embedding/embedding.cpp
index d8eaaa2691..33ef2a7521 100644
--- a/examples/embedding/embedding.cpp
+++ b/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)) {
diff --git a/examples/eval-callback/eval-callback.cpp b/examples/eval-callback/eval-callback.cpp
index bd58734979..17d162d95d 100644
--- a/examples/eval-callback/eval-callback.cpp
+++ b/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;
diff --git a/examples/gen-docs/gen-docs.cpp b/examples/gen-docs/gen-docs.cpp
index 0aa33e8245..7ba7d79f72 100644
--- a/examples/gen-docs/gen-docs.cpp
+++ b/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()) {
diff --git a/examples/gguf-hash/gguf-hash.cpp b/examples/gguf-hash/gguf-hash.cpp
index 9523ec122f..331de301ff 100644
--- a/examples/gguf-hash/gguf-hash.cpp
+++ b/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 <stdexcept>
-#include <algorithm>
+#include <cstdlib>   /* abort() */
 #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);
diff --git a/examples/gguf/gguf.cpp b/examples/gguf/gguf.cpp
index 499cfacc92..79ad38711e 100644
--- a/examples/gguf/gguf.cpp
+++ b/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");
diff --git a/examples/lookahead/lookahead.cpp b/examples/lookahead/lookahead.cpp
index aa6efa62b3..d5fde081c5 100644
--- a/examples/lookahead/lookahead.cpp
+++ b/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)) {
diff --git a/examples/lookup/lookup-create.cpp b/examples/lookup/lookup-create.cpp
index f7b6ea1b19..439e3f726e 100644
--- a/examples/lookup/lookup-create.cpp
+++ b/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)) {
diff --git a/examples/lookup/lookup-merge.cpp b/examples/lookup/lookup-merge.cpp
index 6871c0f5fd..ee3c7249cf 100644
--- a/examples/lookup/lookup-merge.cpp
+++ b/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);
diff --git a/examples/lookup/lookup-stats.cpp b/examples/lookup/lookup-stats.cpp
index ae28b2e6e8..c3158281c7 100644
--- a/examples/lookup/lookup-stats.cpp
+++ b/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)) {
diff --git a/examples/lookup/lookup.cpp b/examples/lookup/lookup.cpp
index c7552ddde1..bd216035c0 100644
--- a/examples/lookup/lookup.cpp
+++ b/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)) {
diff --git a/examples/parallel/parallel.cpp b/examples/parallel/parallel.cpp
index c92173ae29..1700ceefbf 100644
--- a/examples/parallel/parallel.cpp
+++ b/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;
diff --git a/examples/passkey/passkey.cpp b/examples/passkey/passkey.cpp
index 8a4faa383b..665191047a 100644
--- a/examples/passkey/passkey.cpp
+++ b/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;
diff --git a/examples/retrieval/retrieval.cpp b/examples/retrieval/retrieval.cpp
index 3f2afd4346..9e05fc2233 100644
--- a/examples/retrieval/retrieval.cpp
+++ b/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)) {
diff --git a/examples/save-load-state/save-load-state.cpp b/examples/save-load-state/save-load-state.cpp
index 5e35dcd603..174c8c7585 100644
--- a/examples/save-load-state/save-load-state.cpp
+++ b/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";
diff --git a/examples/simple-chat/simple-chat.cpp b/examples/simple-chat/simple-chat.cpp
index 57195df331..97e9dc9842 100644
--- a/examples/simple-chat/simple-chat.cpp
+++ b/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;
diff --git a/examples/simple/simple.cpp b/examples/simple/simple.cpp
index d09771d104..9f0a25d713 100644
--- a/examples/simple/simple.cpp
+++ b/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
diff --git a/examples/speculative-simple/speculative-simple.cpp b/examples/speculative-simple/speculative-simple.cpp
index d8b1f5a480..8a1cbd96c2 100644
--- a/examples/speculative-simple/speculative-simple.cpp
+++ b/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)) {
diff --git a/examples/speculative/speculative.cpp b/examples/speculative/speculative.cpp
index 3e5cf5f46b..250c5b7c62 100644
--- a/examples/speculative/speculative.cpp
+++ b/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
diff --git a/examples/sycl/ls-sycl-device.cpp b/examples/sycl/ls-sycl-device.cpp
index 74a8b7fd81..3bdc405982 100644
--- a/examples/sycl/ls-sycl-device.cpp
+++ b/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;
 }
diff --git a/examples/training/finetune.cpp b/examples/training/finetune.cpp
index c82de8d35d..e20f89488f 100644
--- a/examples/training/finetune.cpp
+++ b/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;
 
diff --git a/tools/batched-bench/batched-bench.cpp b/tools/batched-bench/batched-bench.cpp
index 0f627c5ff6..224f0e1f1c 100644
--- a/tools/batched-bench/batched-bench.cpp
+++ b/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)) {
diff --git a/tools/completion/completion.cpp b/tools/completion/completion.cpp
index e13816f735..2e0f087184 100644
--- a/tools/completion/completion.cpp
+++ b/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;
 
diff --git a/tools/cvector-generator/cvector-generator.cpp b/tools/cvector-generator/cvector-generator.cpp
index 3ba7c52950..4c8ca61ec4 100644
--- a/tools/cvector-generator/cvector-generator.cpp
+++ b/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";
diff --git a/tools/export-lora/export-lora.cpp b/tools/export-lora/export-lora.cpp
index 41f426208f..50774c59ba 100644
--- a/tools/export-lora/export-lora.cpp
+++ b/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";
diff --git a/tools/gguf-split/gguf-split.cpp b/tools/gguf-split/gguf-split.cpp
index 30e771564e..f99f0299b9 100644
--- a/tools/gguf-split/gguf-split.cpp
+++ b/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);
 
diff --git a/tools/imatrix/imatrix.cpp b/tools/imatrix/imatrix.cpp
index e025c114b4..bbedb159cd 100644
--- a/tools/imatrix/imatrix.cpp
+++ b/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";
diff --git a/tools/llama-bench/llama-bench.cpp b/tools/llama-bench/llama-bench.cpp
index 7da6c3957c..7a75026550 100644
--- a/tools/llama-bench/llama-bench.cpp
+++ b/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");
diff --git a/tools/mtmd/deprecation-warning.cpp b/tools/mtmd/deprecation-warning.cpp
index dded0a56af..2b31a9d8b0 100644
--- a/tools/mtmd/deprecation-warning.cpp
+++ b/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];
diff --git a/tools/mtmd/mtmd-cli.cpp b/tools/mtmd/mtmd-cli.cpp
index 054c7faa6a..09b8e311c5 100644
--- a/tools/mtmd/mtmd-cli.cpp
+++ b/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;
diff --git a/tools/perplexity/perplexity.cpp b/tools/perplexity/perplexity.cpp
index 433b747f0d..cc5ea99c4d 100644
--- a/tools/perplexity/perplexity.cpp
+++ b/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;
diff --git a/tools/quantize/quantize.cpp b/tools/quantize/quantize.cpp
index 59bf9bd3fd..0a483328ee 100644
--- a/tools/quantize/quantize.cpp
+++ b/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]);
     }
diff --git a/tools/rpc/rpc-server.cpp b/tools/rpc/rpc-server.cpp
index 6feb0e91f3..03ab78e5f0 100644
--- a/tools/rpc/rpc-server.cpp
+++ b/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;
diff --git a/tools/server/server.cpp b/tools/server/server.cpp
index f353dcdde7..fab0bb587f 100644
--- a/tools/server/server.cpp
+++ b/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;
 
diff --git a/tools/tokenize/tokenize.cpp b/tools/tokenize/tokenize.cpp
index 7375759ebe..32cf8c8eb9 100644
--- a/tools/tokenize/tokenize.cpp
+++ b/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();
 
diff --git a/tools/tts/tts.cpp b/tools/tts/tts.cpp
index ac55a8b1ca..dc2fa494b8 100644
--- a/tools/tts/tts.cpp
+++ b/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";