Support multimodal on /completions endpoint

- If the 'multimodal_data' field is present, treat each entry in the list as a base64 encoded multimodal data, and append it to the files passed to the completion handler
- Just check for mtmd support, and not only openai-emulation, in the completion handler implementation.

Author: 65a

tools/server/server.cpp

@@ -4250,7 +4250,7 @@ int main(int argc, char ** argv) {
             // process prompt
             std::vector<server_tokens> inputs;
 
-            if (oaicompat && has_mtmd) {
+            if (has_mtmd) {
                 // multimodal
                 std::string prompt_str = prompt.get<std::string>();
                 mtmd_input_text inp_txt = {
@@ -4369,7 +4369,12 @@ int main(int argc, char ** argv) {
 
     const auto handle_completions = [&handle_completions_impl](const httplib::Request & req, httplib::Response & res) {
         json data = json::parse(req.body);
-        std::vector<raw_buffer> files; // dummy
+        std::vector<raw_buffer> files;
+        if (data.find("multimodal_data") != data.end()) {
+            for (const auto& entry : data.at("multimodal_data")) {
+                files.push_back(base64_decode(entry));
+            }
+        }
         handle_completions_impl(
             SERVER_TASK_TYPE_COMPLETION,
             data,