diff --git a/extensions/api/util.py b/extensions/api/util.py
index 206f259..2c7e73f 100644
--- a/extensions/api/util.py
+++ b/extensions/api/util.py
@@ -74,8 +74,8 @@ def build_parameters(body, chat=False):
         if str(character) == "None":
             character = "Assistant"
 
-        name1, name2, _, greeting, context, _ = load_character_memoized(character, str(body.get('your_name', shared.settings['name1'])), '', instruct=False)
-        name1_instruct, name2_instruct, _, _, context_instruct, turn_template = load_character_memoized(instruction_template, '', '', instruct=True)
+        name1, name2, _, greeting, context, _, _ = load_character_memoized(character, str(body.get('your_name', shared.settings['name1'])), '', instruct=False)
+        name1_instruct, name2_instruct, _, _, context_instruct, turn_template, _ = load_character_memoized(instruction_template, '', '', instruct=True)
         generate_params.update({
             'mode': str(body.get('mode', 'chat')),
             'name1': str(body.get('name1', name1)),
diff --git a/extensions/openai/script.py b/extensions/openai/script.py
index 94e4160..40574a7 100644
--- a/extensions/openai/script.py
+++ b/extensions/openai/script.py
@@ -91,6 +91,10 @@ async def openai_completions(request: Request, request_data: CompletionRequest):
             async with streaming_semaphore:
                 response = OAIcompletions.stream_completions(to_dict(request_data), is_legacy=is_legacy)
                 for resp in response:
+                    disconnected = await request.is_disconnected()
+                    if disconnected:
+                        break
+
                     yield {"data": json.dumps(resp)}
 
         return EventSourceResponse(generator())  # SSE streaming
@@ -110,6 +114,10 @@ async def openai_chat_completions(request: Request, request_data: ChatCompletion
             async with streaming_semaphore:
                 response = OAIcompletions.stream_chat_completions(to_dict(request_data), is_legacy=is_legacy)
                 for resp in response:
+                    disconnected = await request.is_disconnected()
+                    if disconnected:
+                        break
+
                     yield {"data": json.dumps(resp)}
 
         return EventSourceResponse(generator())  # SSE streaming