server : support preserving reasoning_content in assistant message (#18994)

* support reasoning_content input

* report template caps to webui

* add docs

* rm commented code

tools/server/README.md

@@ -781,6 +781,7 @@ By default, it is read-only. To make POST request to change global properties, y
   "total_slots": 1,
   "model_path": "../models/Meta-Llama-3.1-8B-Instruct-Q4_K_M.gguf",
   "chat_template": "...",
+  "chat_template_caps": {},
   "modalities": {
     "vision": false
   },
@@ -793,6 +794,7 @@ By default, it is read-only. To make POST request to change global properties, y
 - `total_slots` - the total number of slots for process requests (defined by `--parallel` option)
 - `model_path` - the path to model file (same with `-m` argument)
 - `chat_template` - the model's original Jinja2 prompt template
+- `chat_template_caps` - capabilities of the chat template (see `common/jinja/caps.h` for more info)
 - `modalities` - the list of supported modalities
 - `is_sleeping` - sleeping status, see [Sleeping on idle](#sleeping-on-idle)
 
@@ -1267,6 +1269,12 @@ This provides information on the performance of the server. It also allows calcu
 
 The total number of tokens in context is equal to `prompt_n + cache_n + predicted_n`
 
+*Reasoning support*
+
+The server supports parsing and returning reasoning via the `reasoning_content` field, similar to Deepseek API.
+
+Reasoning input (preserve reasoning in history) is also supported by some specific templates. For more details, please refer to [PR#18994](https://github.com/ggml-org/llama.cpp/pull/18994).
+
 ### POST `/v1/responses`: OpenAI-compatible Responses API
 
 *Options:*

tools/server/server-context.cpp

@@ -2903,6 +2903,7 @@ server_context_meta server_context::get_meta() const {
         /* pooling_type           */ llama_pooling_type(impl->ctx),
 
         /* chat_params            */ impl->chat_params,
+        /* chat_template_caps     */ common_chat_templates_get_caps(impl->chat_params.tmpls.get()),
 
         /* bos_token_str          */ bos_token_str,
         /* eos_token_str          */ eos_token_str,
@@ -3410,6 +3411,7 @@ void server_routes::init_routes() {
             { "webui",                       params.webui },
             { "webui_settings",              meta->json_webui_settings },
             { "chat_template",               tmpl_default },
+            { "chat_template_caps",          meta->chat_template_caps },
             { "bos_token",                   meta->bos_token_str },
             { "eos_token",                   meta->eos_token_str },
             { "build_info",                  meta->build_info },

tools/server/server-context.h

@@ -22,6 +22,7 @@ struct server_context_meta {
 
     // chat params
     server_chat_params & chat_params;
+    std::map<std::string, bool> chat_template_caps;
 
     // tokens
     std::string bos_token_str;

tools/server/server-task.cpp

@@ -700,7 +700,7 @@ json server_task_result_cmpl_final::to_json_oaicompat_chat() {
     json choice {
         {"finish_reason", finish_reason},
         {"index", index},
-        {"message", msg.to_json_oaicompat<json>()},
+        {"message", msg.to_json_oaicompat()},
     };
 
     if (!stream && probs_output.size() > 0) {
@@ -750,7 +750,7 @@ json server_task_result_cmpl_final::to_json_oaicompat_chat_stream() {
                 json {
                     {"finish_reason", nullptr},
                     {"index", 0},
-                    {"delta", common_chat_msg_diff_to_json_oaicompat<json>(diff)},
+                    {"delta", common_chat_msg_diff_to_json_oaicompat(diff)},
                 },
             })},
             {"created", t},
@@ -1383,7 +1383,7 @@ json server_task_result_cmpl_partial::to_json_oaicompat_chat() {
     }
 
     for (const auto & diff : oaicompat_msg_diffs) {
-        add_delta(common_chat_msg_diff_to_json_oaicompat<json>(diff));
+        add_delta(common_chat_msg_diff_to_json_oaicompat(diff));
     }
 
     if (!deltas.empty()) {