common: allow the warmup to be disabled in llama_init_from_gpt_params

common/common.cpp

@@ -2194,7 +2194,7 @@ std::tuple<struct llama_model *, struct llama_context *> llama_init_from_gpt_par
         params.sparams.logit_bias[llama_token_eos(model)] = -INFINITY;
     }
 
-    {
+    if (params.warmup) {
         LOG("warming up the model with an empty run\n");
 
         std::vector<llama_token> tmp = { llama_token_bos(model), llama_token_eos(model), };

common/common.h

@@ -159,6 +159,7 @@ struct gpt_params {
     bool infill            = false; // use infill mode
     bool dump_kv_cache     = false; // dump the KV cache contents for debugging purposes
     bool no_kv_offload     = false; // disable KV offloading
+    bool warmup            = true;  // warmup run
 
     std::string cache_type_k = "f16"; // KV cache data type for the K
     std::string cache_type_v = "f16"; // KV cache data type for the V

examples/ggml-debug/ggml-debug.cpp

@@ -136,6 +136,7 @@ int main(int argc, char ** argv) {
     // it will be executed for each node during the graph computation
     params.cb_eval = ggml_debug;
     params.cb_eval_user_data = &cb_data;
+    params.warmup = false;
 
     // init
     llama_model * model;

examples/imatrix/imatrix.cpp

@@ -601,6 +601,7 @@ int main(int argc, char ** argv) {
     // it will be executed for each node during the graph computation
     params.cb_eval = ik_collect_imatrix;
     params.cb_eval_user_data = NULL;
+    params.warmup = false;
 
     // init
     llama_model * model;