initial commit for handling extra template kwargs

Author: matteoserva

common/chat.cpp

@@ -34,6 +34,7 @@ struct templates_params {
     bool add_generation_prompt = true;
     bool extract_reasoning     = true;
     std::chrono::system_clock::time_point now = std::chrono::system_clock::now();
+    json extra_context;
 };
 
 common_chat_tool_choice common_chat_tool_choice_parse_oaicompat(const std::string & tool_choice) {
@@ -1582,7 +1583,7 @@ static common_chat_msg common_chat_parse_hermes_2_pro(const std::string& input,
 
 static common_chat_params common_chat_params_init_without_tools(const common_chat_template & tmpl, const struct templates_params & inputs) {
     common_chat_params data;
-    data.prompt = apply(tmpl, inputs.messages, inputs.tools.empty() ? json() : inputs.tools, inputs.add_generation_prompt);
+    data.prompt = apply(tmpl, inputs.messages, inputs.tools.empty() ? json() : inputs.tools, inputs.add_generation_prompt,inputs.extra_context);
     data.format = COMMON_CHAT_FORMAT_CONTENT_ONLY;
     data.grammar_lazy = false;
     if (!inputs.json_schema.is_null()) {
@@ -1613,6 +1614,12 @@ static common_chat_params common_chat_templates_apply_jinja(
     params.tool_choice = inputs.tool_choice;
     params.grammar = inputs.grammar;
     params.now = inputs.now;
+
+    for(auto el: inputs.chat_template_kwargs)
+    {
+        params.extra_context[el.first] = json::parse(el.second);
+    }
+
     if (!inputs.json_schema.empty()) {
         params.json_schema = json::parse(inputs.json_schema);
     }

common/chat.h

@@ -6,6 +6,7 @@
 #include <chrono>
 #include <string>
 #include <vector>
+#include <map>
 
 struct common_chat_templates;
 
@@ -73,6 +74,7 @@ struct common_chat_templates_inputs {
     bool parallel_tool_calls = false;
     bool extract_reasoning     = true;
     std::chrono::system_clock::time_point now = std::chrono::system_clock::now();
+    std::map<std::string,std::string> chat_template_kwargs;
 };
 
 struct common_chat_params {

tools/server/utils.hpp

@@ -726,6 +726,12 @@ static json oaicompat_completion_params_parse(
     inputs.parallel_tool_calls   = json_value(body, "parallel_tool_calls", false);
     inputs.extract_reasoning     = reasoning_format != COMMON_REASONING_FORMAT_NONE;
     inputs.add_generation_prompt = json_value(body, "add_generation_prompt", true);
+
+    auto chat_template_kwargs_object = json_value(body, "chat_template_kwargs", json::object());
+    for (const auto & item : chat_template_kwargs_object.items()) {
+        inputs.chat_template_kwargs[item.key()] = item.value().dump();
+    }
+
     if (!inputs.tools.empty() && inputs.tool_choice != COMMON_CHAT_TOOL_CHOICE_NONE && body.contains("grammar")) {
         throw std::runtime_error("Cannot use custom grammar constraints with tools.");
     }