From 61b63705dcf39997f2c934ce59b8bde5d1adfbd6 Mon Sep 17 00:00:00 2001
From: Georgi Gerganov <ggerganov@gmail.com>
Date: Tue, 5 Mar 2024 19:55:19 +0200
Subject: [PATCH] server : refactor system prompt update at start

---
 examples/server/server.cpp | 38 +++++++++++++++++---------------------
 examples/server/utils.hpp  | 11 +++++++++++
 2 files changed, 28 insertions(+), 21 deletions(-)

diff --git a/examples/server/server.cpp b/examples/server/server.cpp
index a092ee61c..77b5321e9 100644
--- a/examples/server/server.cpp
+++ b/examples/server/server.cpp
@@ -115,6 +115,7 @@ struct server_params {
     std::string hostname      = "127.0.0.1";
     std::string public_path   = "examples/server/public";
     std::string chat_template = "";
+    std::string system_prompt = "";
 
     std::vector<std::string> api_keys;
 
@@ -1024,7 +1025,7 @@ struct llama_server_context {
         system_need_update = false;
     }
 
-    void system_prompt_process(const json & sys_props) {
+    void system_prompt_set(const json & sys_props) {
         system_prompt  = sys_props.value("prompt", "");
         name_user      = sys_props.value("anti_prompt", "");
         name_assistant = sys_props.value("assistant_name", "");
@@ -1418,7 +1419,7 @@ struct llama_server_context {
                     }
 
                     if (task.data.contains("system_prompt")) {
-                        system_prompt_process(task.data["system_prompt"]);
+                        system_prompt_set(task.data["system_prompt"]);
 
                         // reset cache_tokens for all slots
                         for (server_slot & slot : slots) {
@@ -1974,7 +1975,10 @@ struct llama_server_context {
                 }
 
                 for (size_t i = 0; i < std::min(cur_p.size, (size_t) n_probs); ++i) {
-                    result.probs.push_back({cur_p.data[i].id, cur_p.data[i].p});
+                    result.probs.push_back({
+                        cur_p.data[i].id,
+                        cur_p.data[i].p
+                    });
                 }
 
                 if (!process_token(result, slot)) {
@@ -2088,7 +2092,7 @@ static void server_print_usage(const char * argv0, const gpt_params & params, co
     printf("\n");
 }
 
-static void server_params_parse(int argc, char ** argv, server_params & sparams, gpt_params & params, llama_server_context & llama) {
+static void server_params_parse(int argc, char ** argv, server_params & sparams, gpt_params & params) {
     gpt_params    default_params;
     server_params default_sparams;
 
@@ -2402,13 +2406,13 @@ static void server_params_parse(int argc, char ** argv, server_params & sparams,
                 invalid_param = true;
                 break;
             }
-            std::string systm_content;
+            std::string system_prompt;
             std::copy(
                 std::istreambuf_iterator<char>(file),
                 std::istreambuf_iterator<char>(),
-                std::back_inserter(systm_content)
+                std::back_inserter(system_prompt)
             );
-            llama.system_prompt_process(json::parse(systm_content));
+            sparams.system_prompt = system_prompt;
         } else if (arg == "-ctk" || arg == "--cache-type-k") {
             params.cache_type_k = argv[++i];
         } else if (arg == "-ctv" || arg == "--cache-type-v") {
@@ -2506,18 +2510,6 @@ static void server_params_parse(int argc, char ** argv, server_params & sparams,
     }
 }
 
-static json format_tokenizer_response(const std::vector<llama_token> & tokens) {
-    return json {
-        {"tokens", tokens}
-    };
-}
-
-static json format_detokenized_response(const std::string & content) {
-    return json {
-        {"content", content}
-    };
-}
-
 static void log_server_request(const httplib::Request & req, const httplib::Response & res) {
     // skip GH copilot requests when using default port
     if (req.path == "/v1/health" || req.path == "/v1/completions") {
@@ -2558,13 +2550,17 @@ int main(int argc, char ** argv) {
     log_disable();
 #endif
     // own arguments required by this example
-    gpt_params params;
+    gpt_params    params;
     server_params sparams;
 
     // struct that contains llama context and inference
     llama_server_context llama;
 
-    server_params_parse(argc, argv, sparams, params, llama);
+    server_params_parse(argc, argv, sparams, params);
+
+    if (!sparams.system_prompt.empty()) {
+        llama.system_prompt_set(json::parse(sparams.system_prompt));
+    }
 
     if (params.model_alias == "unknown") {
         params.model_alias = params.model;
diff --git a/examples/server/utils.hpp b/examples/server/utils.hpp
index 7e08209fc..df0a27782 100644
--- a/examples/server/utils.hpp
+++ b/examples/server/utils.hpp
@@ -531,3 +531,14 @@ static json format_embeddings_response_oaicompat(const json & request, const jso
     return res;
 }
 
+static json format_tokenizer_response(const std::vector<llama_token> & tokens) {
+    return json {
+        {"tokens", tokens}
+    };
+}
+
+static json format_detokenized_response(const std::string & content) {
+    return json {
+        {"content", content}
+    };
+}