fix: Use '\n' seperator for EventSourceResponse (#1188)

this fixes compatibility with some OpenAI clients, including BetterChatGPT (ztjhz/BetterChatGPT#537).

Co-authored-by: Andrei <[email protected]>

Author: khimaros

llama_cpp/server/app.py

@@ -290,6 +290,7 @@ def iterator() -> Iterator[llama_cpp.CreateCompletionStreamResponse]:
                 inner_send_chan=send_chan,
                 iterator=iterator(),
             ),
+            sep='\n',
         )
     else:
         return iterator_or_completion
@@ -382,6 +383,7 @@ def iterator() -> Iterator[llama_cpp.ChatCompletionChunk]:
                 inner_send_chan=send_chan,
                 iterator=iterator(),
             ),
+            sep='\n',
         )
     else:
         return iterator_or_completion