diff --git a/detector/server.py b/detector/server.py
index 90cdcf8..34a0c85 100644
--- a/detector/server.py
+++ b/detector/server.py
@@ -68,7 +68,7 @@ def serve_forever(server, model, tokenizer, device):
     globals()['tokenizer'] = tokenizer
     globals()['device'] = device
 
-    log('Ready to serve')
+    log(f'Ready to serve at http://localhost:{server.server_address[1]}')
     server.serve_forever()
 
 
@@ -93,7 +93,7 @@ def main(checkpoint, port=8080, device='cuda' if torch.cuda.is_available() else
     server = HTTPServer(('0.0.0.0', port), RequestHandler)
 
     # avoid calling CUDA API before forking; doing so in a subprocess is fine.
-    num_workers = int(subprocess.check_output(['python', '-c', 'import torch; print(torch.cuda.device_count())']))
+    num_workers = int(subprocess.check_output([sys.executable, '-c', 'import torch; print(torch.cuda.device_count())']))
 
     if num_workers <= 1:
         serve_forever(server, model, tokenizer, device)
diff --git a/detector/train.py b/detector/train.py
index ff8d79a..748ef4d 100644
--- a/detector/train.py
+++ b/detector/train.py
@@ -281,7 +281,7 @@ if __name__ == '__main__':
     parser.add_argument('--weight-decay', type=float, default=0)
     args = parser.parse_args()
 
-    nproc = int(subprocess.check_output(['python', '-c', "import torch;"
+    nproc = int(subprocess.check_output([sys.executable, '-c', "import torch;"
                                          "print(torch.cuda.device_count() if torch.cuda.is_available() else 1)"]))
     if nproc > 1:
         print(f'Launching {nproc} processes ...', file=sys.stderr)