nits

tools/server/server-context.cpp

@@ -694,7 +694,7 @@ struct server_slot {
 // caller need to update prompt.tokens after a successful call to keep track of the processing progress
 // note: this is not a member of server_slot because we want to run it inside yield_to_queue
 //       slot is passed as const to avoid accidental modification of the slot state
-//       only llama_context / mtmd_context / mbatch are allowed to be used inside
+//       some pointers are allowed to be used, they are not used by to_json()
 static int process_mtmd_chunk(const server_slot & slot, mtmd::batch_ptr & mbatch, size_t idx, size_t & n_tokens_out) {
     GGML_ASSERT(slot.mctx);
     const auto & mctx = slot.mctx;

tools/server/server-queue.cpp

@@ -192,7 +192,7 @@ void server_queue::worker_stop() {
 }
 
 void server_queue::yield_to_queue(std::function<void()> && work) {
-    GGML_ASSERT(worker.thread.joinable() && "yield_to_queue() must be called from the start_loop() thread");
+    GGML_ASSERT(worker.thread.joinable() && "yield_to_queue() requires start_loop() to be running");
 
     QUE_DBG("%s", "yielding to queue\n");
 

tools/server/server-queue.h

@@ -4,6 +4,7 @@
 
 #include <condition_variable>
 #include <deque>
+#include <exception>
 #include <mutex>
 #include <thread>
 #include <vector>