Merge remote-tracking branch 'duncan/api-stream-double-write-fix' into concedo_experimental

koboldcpp.py

@@ -572,12 +572,14 @@ class ServerRequestHandler(http.server.SimpleHTTPRequestHandler):
                 newprompt = fullprompt
 
                 gen = asyncio.run(self.handle_request(genparams, newprompt, basic_api_flag, kai_sse_stream_flag))
-                try:
+
-                    self.send_response(200)
+                if not kai_sse_stream_flag:
-                    self.end_headers()
+                    try:
-                    self.wfile.write(json.dumps(gen).encode())
+                        self.send_response(200)
-                except:
+                        self.end_headers()
-                    print("Generate: The response could not be sent, maybe connection was terminated?")
+                        self.wfile.write(json.dumps(gen).encode())
+                    except:
+                        print("Generate: The response could not be sent, maybe connection was terminated?")
 
                 return
         finally: