chore(deps): update llama.cpp and sync with upstream changes (#4950)

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

Makefile

@@ -6,7 +6,7 @@ BINARY_NAME=local-ai
 DETECT_LIBS?=true
 
 # llama.cpp versions
-CPPLLAMA_VERSION?=5bbe6a9fe9a8796a9389c85accec89dbc4d91e39
+CPPLLAMA_VERSION?=5e43f104cca1a14874e980326a506b44fde022b8
 
 # whisper.cpp version
 WHISPER_REPO?=https://github.com/ggerganov/whisper.cpp

backend/backend.proto

@@ -165,7 +165,6 @@ message Reply {
 
 message GrammarTrigger {
   string word = 1;
-  bool at_start = 2;
 }
 
 message ModelOptions {

backend/cpp/llama/grpc-server.cpp

@@ -469,7 +469,7 @@ struct llama_server_context
     bool has_eos_token      = true;
 
     bool grammar_lazy = false;
-    std::vector<common_grammar_trigger> grammar_trigger_words;
+    std::vector<common_grammar_trigger> grammar_triggers;
 
     int32_t n_ctx;  // total context for all clients / slots
 
@@ -709,7 +709,7 @@ struct llama_server_context
         slot->sparams.grammar           = json_value(data, "grammar",           default_sparams.grammar);
         slot->sparams.n_probs           = json_value(data, "n_probs",           default_sparams.n_probs);
         slot->sparams.min_keep          = json_value(data, "min_keep",          default_sparams.min_keep);
-        slot->sparams.grammar_trigger_words = grammar_trigger_words;
+        slot->sparams.grammar_triggers = grammar_triggers;
         slot->sparams.grammar_lazy = grammar_lazy;
 
         if (slot->n_predict > 0 && slot->params.n_predict > slot->n_predict) {
@@ -2393,12 +2393,12 @@ static void params_parse(const backend::ModelOptions* request,
         llama.grammar_lazy = true;
         for (int i = 0; i < request->grammartriggers_size(); i++) {
             common_grammar_trigger trigger;
-            trigger.word = request->grammartriggers(i).word();
-            trigger.at_start = request->grammartriggers(i).at_start();
-            llama.grammar_trigger_words.push_back(trigger);
+	    trigger.type = COMMON_GRAMMAR_TRIGGER_TYPE_WORD;
+            trigger.value = request->grammartriggers(i).word();
+	    // trigger.at_start = request->grammartriggers(i).at_start();
+            llama.grammar_triggers.push_back(trigger);
             LOG_INFO("grammar trigger", {
-                { "word", trigger.word },
-                { "at_start", trigger.at_start }
+                { "word", trigger.value },
             });
         }
     }

core/backend/options.go

@@ -122,7 +122,6 @@ func grpcModelOpts(c config.BackendConfig) *pb.ModelOptions {
 	for _, t := range c.FunctionsConfig.GrammarConfig.GrammarTriggers {
 		triggers = append(triggers, &pb.GrammarTrigger{
 			Word:    t.Word,
-			AtStart: t.AtStart,
 		})
 
 	}

pkg/functions/parse.go

@@ -54,7 +54,6 @@ type GrammarConfig struct {
 type GrammarTrigger struct {
 	// Trigger is the string that triggers the grammar
 	Word    string `yaml:"word"`
-	AtStart bool   `yaml:"at_start"`
 }
 
 // FunctionsConfig is the configuration for the tool/function call.