ui fix

src/core/engine.py

@@ -59,32 +59,49 @@ class ModelEngine:
         if kwargs.get("tool_choice"):
             llama_kwargs["tool_choice"] = kwargs["tool_choice"]
 
-        acquired = False
-        while not acquired:
-            if abort_event and abort_event.is_set():
-                print("Request aborted while waiting in queue.")
-                return
-            acquired = self.lock.acquire(timeout=0.5)
+        # Если включен стриминг, выносим yield во внутреннюю функцию-генератор
+        if stream_mode:
 
-        try:
-            if abort_event and abort_event.is_set():
-                return
-
-            response = self.llm.create_chat_completion(**llama_kwargs)
+            def stream_generator():
+                acquired = False
+                while not acquired:
+                    if abort_event and abort_event.is_set():
+                        print("Request aborted while waiting in queue.")
+                        return
+                    acquired = self.lock.acquire(timeout=0.5)
 
-            # Если просят стрим - отдаем генератор
-            if stream_mode:
-                for chunk in response:
+                try:
                     if abort_event and abort_event.is_set():
-                        print("Request aborted during generation.")
-                        break
-                    yield chunk
-            # Если просят целый ответ (Агенты) - отдаем сразу словарь
-            else:
+                        return
+
+                    response = self.llm.create_chat_completion(**llama_kwargs)
+                    for chunk in response:
+                        if abort_event and abort_event.is_set():
+                            print("Request aborted during generation.")
+                            break
+                        yield chunk
+                finally:
+                    self.lock.release()
+
+            return stream_generator()
+
+        # Если стриминг выключен (режим Агента), возвращаем обычный словарь
+        else:
+            acquired = False
+            while not acquired:
+                if abort_event and abort_event.is_set():
+                    print("Request aborted while waiting in queue.")
+                    return
+                acquired = self.lock.acquire(timeout=0.5)
+
+            try:
+                if abort_event and abort_event.is_set():
+                    return
+
+                response = self.llm.create_chat_completion(**llama_kwargs)
                 return response
-
-        finally:
-            self.lock.release()
+            finally:
+                self.lock.release()
 
 
 engine = ModelEngine()

src/ui/components.py

@@ -49,9 +49,7 @@ def create_ui():
 
             # Основная область чата
             with gr.Column(elem_id="col-chat-main"):
-                chatbot = gr.Chatbot(
-                    elem_id="chatbot", show_label=False, bubble_full_width=False
-                )
+                chatbot = gr.Chatbot(elem_id="chatbot", show_label=False)
 
                 with gr.Row(elem_id="input-area"):
                     msg = gr.Textbox(