examples : move gpt_init() after parsing the cli args

examples/batched-bench/batched-bench.cpp

@@ -15,14 +15,14 @@ static void print_usage(int, char ** argv) {
 }
 
 int main(int argc, char ** argv) {
-    gpt_init();
-
     gpt_params params;
 
     if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_BENCH, print_usage)) {
         return 1;
     }
 
+    gpt_init();
+
     int is_pp_shared = params.is_pp_shared;
 
     std::vector<int> n_pp = params.n_pp;

examples/batched/batched.cpp

@@ -15,8 +15,6 @@ static void print_usage(int, char ** argv) {
 }
 
 int main(int argc, char ** argv) {
-    gpt_log_init();
-
     gpt_params params;
 
     params.prompt = "Hello my name is";
@@ -26,6 +24,7 @@ int main(int argc, char ** argv) {
         return 1;
     }
 
+    gpt_init();
 
     // number of parallel batches
     int n_parallel = params.n_parallel;

examples/embedding/embedding.cpp

@@ -79,14 +79,14 @@ static void batch_decode(llama_context * ctx, llama_batch & batch, float * outpu
 }
 
 int main(int argc, char ** argv) {
-    gpt_log_init();
-
     gpt_params params;
 
     if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_EMBEDDING)) {
         return 1;
     }
 
+    gpt_init();
+
     params.embedding = true;
     // For non-causal models, batch size must be equal to ubatch size
     params.n_ubatch = params.n_batch;

examples/eval-callback/eval-callback.cpp

@@ -140,8 +140,6 @@ static bool run(llama_context * ctx, const gpt_params & params) {
 }
 
 int main(int argc, char ** argv) {
-    gpt_init();
-
     callback_data cb_data;
 
     gpt_params params;
@@ -150,6 +148,8 @@ int main(int argc, char ** argv) {
         return 1;
     }
 
+    gpt_init();
+
     llama_backend_init();
     llama_numa_init(params.numa);
 

examples/gritlm/gritlm.cpp

@@ -158,6 +158,8 @@ int main(int argc, char * argv[]) {
         return 1;
     }
 
+    gpt_init();
+
     llama_model_params mparams = llama_model_params_from_gpt_params(params);
     llama_context_params cparams = llama_context_params_from_gpt_params(params);
 

examples/imatrix/imatrix.cpp

@@ -573,8 +573,6 @@ static bool compute_imatrix(llama_context * ctx, const gpt_params & params) {
 }
 
 int main(int argc, char ** argv) {
-    gpt_init();
-
     gpt_params params;
 
     params.n_ctx = 512;
@@ -585,6 +583,8 @@ int main(int argc, char ** argv) {
         return 1;
     }
 
+    gpt_init();
+
     params.n_batch = std::min(params.n_batch, params.n_ctx);
 
     g_collector.set_params(params);

examples/infill/infill.cpp

@@ -104,8 +104,6 @@ static void sigint_handler(int signo) {
 #endif
 
 int main(int argc, char ** argv) {
-    gpt_init();
-
     gpt_params params;
     g_params = &params;
 
@@ -113,6 +111,8 @@ int main(int argc, char ** argv) {
         return 1;
     }
 
+    gpt_init();
+
     auto & sparams = params.sparams;
 
     console::init(params.simple_io, params.use_color);

examples/llava/llava-cli.cpp

@@ -270,8 +270,6 @@ static void llava_free(struct llava_context * ctx_llava) {
 }
 
 int main(int argc, char ** argv) {
-    gpt_init();
-
     ggml_time_init();
 
     gpt_params params;
@@ -280,6 +278,8 @@ int main(int argc, char ** argv) {
         return 1;
     }
 
+    gpt_init();
+
     if (params.mmproj.empty() || (params.image.empty() && !prompt_contains_image(params.prompt))) {
         print_usage(argc, argv);
         return 1;

examples/llava/minicpmv-cli.cpp

@@ -248,8 +248,6 @@ static const char * llama_loop(struct llava_context * ctx_llava,struct gpt_sampl
 }
 
 int main(int argc, char ** argv) {
-    gpt_init();
-
     ggml_time_init();
 
     gpt_params params;
@@ -258,6 +256,8 @@ int main(int argc, char ** argv) {
         return 1;
     }
 
+    gpt_init();
+
     if (params.mmproj.empty() || (params.image.empty())) {
         show_additional_info(argc, argv);
         return 1;

examples/lookahead/lookahead.cpp

@@ -37,14 +37,14 @@ struct ngram_container {
 };
 
 int main(int argc, char ** argv) {
-    gpt_init();
-
     gpt_params params;
 
     if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_COMMON)) {
         return 1;
     }
 
+    gpt_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-stats.cpp

@@ -13,14 +13,14 @@
 #include <vector>
 
 int main(int argc, char ** argv){
-    gpt_init();
-
     gpt_params params;
 
     if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_LOOKUP)) {
         return 1;
     }
 
+    gpt_init();
+
     const int n_draft = params.n_draft;
 
     // init llama.cpp

examples/lookup/lookup.cpp

@@ -13,14 +13,14 @@
 #include <vector>
 
 int main(int argc, char ** argv){
-    gpt_init();
-
     gpt_params params;
 
     if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_LOOKUP)) {
         return 1;
     }
 
+    gpt_init();
+
     // max. number of additional tokens to draft if match is found
     const int n_draft = params.n_draft;
 

examples/main/main.cpp

@@ -131,14 +131,14 @@ static std::string chat_add_and_format(struct llama_model * model, std::vector<l
 }
 
 int main(int argc, char ** argv) {
-    gpt_init();
-
     gpt_params params;
     g_params = &params;
     if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_MAIN, print_usage)) {
         return 1;
     }
 
+    gpt_init();
+
     auto & sparams = params.sparams;
 
     // save choice to use color for later

examples/parallel/parallel.cpp

@@ -103,14 +103,14 @@ static std::vector<std::string> split_string(const std::string& input, char deli
 int main(int argc, char ** argv) {
     srand(1234);
 
-    gpt_init();
-
     gpt_params params;
 
     if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_PARALLEL)) {
         return 1;
     }
 
+    gpt_init();
+
     // number of simultaneous "clients" to simulate
     const int32_t n_clients = params.n_parallel;
 

examples/passkey/passkey.cpp

@@ -15,8 +15,6 @@ static void print_usage(int, char ** argv) {
 }
 
 int main(int argc, char ** argv) {
-    gpt_init();
-
     gpt_params params;
 
     params.n_junk = 250;
@@ -27,6 +25,8 @@ int main(int argc, char ** argv) {
         return 1;
     }
 
+    gpt_init();
+
     int n_junk = params.n_junk;
     int n_keep = params.n_keep;
     int n_grp  = params.grp_attn_n;

examples/perplexity/perplexity.cpp

@@ -1957,8 +1957,6 @@ static void kl_divergence(llama_context * ctx, const gpt_params & params) {
 }
 
 int main(int argc, char ** argv) {
-    gpt_init();
-
     gpt_params params;
 
     params.n_ctx = 512;
@@ -1968,6 +1966,8 @@ int main(int argc, char ** argv) {
         return 1;
     }
 
+    gpt_init();
+
     const int32_t n_ctx = params.n_ctx;
 
     if (n_ctx <= 0) {

examples/retrieval/retrieval.cpp

@@ -112,14 +112,14 @@ static void batch_decode(llama_context * ctx, llama_batch & batch, float * outpu
 }
 
 int main(int argc, char ** argv) {
-    gpt_init();
-
     gpt_params params;
 
     if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_RETRIEVAL, print_usage)) {
         return 1;
     }
 
+    gpt_init();
+
     // For BERT models, batch size must be equal to ubatch size
     params.n_ubatch = params.n_batch;
     params.embedding = true;

examples/server/server.cpp

@@ -2316,8 +2316,6 @@ inline void signal_handler(int signal) {
 }
 
 int main(int argc, char ** argv) {
-    gpt_init();
-
     // own arguments required by this example
     gpt_params params;
 
@@ -2325,6 +2323,8 @@ int main(int argc, char ** argv) {
         return 1;
     }
 
+    gpt_init();
+
     const bool verbose = params.verbosity > 0;
 
     // struct that contains llama context and inference

examples/simple/simple.cpp

@@ -12,8 +12,6 @@ static void print_usage(int, char ** argv) {
 }
 
 int main(int argc, char ** argv) {
-    gpt_init();
-
     gpt_params params;
 
     params.prompt = "Hello my name is";
@@ -23,6 +21,8 @@ int main(int argc, char ** argv) {
         return 1;
     }
 
+    gpt_init();
+
     // total length of the sequence including the prompt
     const int n_predict = params.n_predict;
 

examples/speculative/speculative.cpp

@@ -30,14 +30,14 @@ struct seq_draft {
 };
 
 int main(int argc, char ** argv) {
-    gpt_init();
-
     gpt_params params;
 
     if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_SPECULATIVE)) {
         return 1;
     }
 
+    gpt_init();
+
     if (params.model_draft.empty()) {
         LOG_ERR("%s: --model-draft is required\n", __func__);
         return 1;