diff --git a/projects/online/online/main.py b/projects/online/online/main.py
index debf6659..a7929bd9 100644
--- a/projects/online/online/main.py
+++ b/projects/online/online/main.py
@@ -5,18 +5,15 @@
 from typing import Iterable, List, Optional, Tuple
 
 import torch
-from amplfi.train.architectures.embeddings import ResNet
-from amplfi.train.architectures.flows import InverseAutoregressiveFlow
 from amplfi.train.architectures.flows.base import FlowArchitecture
 from architectures import Architecture
-from ml4gw.nn.norm import GroupNorm1DGetter
 from ml4gw.transforms import ChannelWiseScaler, SpectralDensity, Whiten
 from torch.multiprocessing import Array, Process, Queue
 
 from ledger.events import EventSet
 from online.utils.buffer import InputBuffer, OutputBuffer
 from online.utils.dataloading import data_iterator
-from online.utils.gdb import GdbServer, gracedb_factory
+from online.utils.gdb import GdbServer, authenticate, gracedb_factory
 from online.utils.ngdd import data_iterator as ngdd_data_iterator
 from online.utils.pastro import fit_or_load_pastro
 from online.utils.pe import run_amplfi, skymap_from_samples
@@ -293,7 +290,7 @@ def event_creation_subprocess(
     pastro_queue: Queue,
 ):
     gdb = gracedb_factory(server, outdir)
-    # last_auth = time.time()
+    last_auth = time.time()
     while True:
         try:
             event = event_queue.get_nowait()
@@ -320,9 +317,9 @@ def event_creation_subprocess(
             # Re-authenticate every 1000 seconds so that
             # the scitoken doesn't expire. Doing it in this
             # loop as it's the earliest point of submission
-            # if last_auth - time.time() > 1000:
-            #     authenticate()
-            #     last_auth = time.time()
+            if last_auth - time.time() > 1000:
+                authenticate()
+                last_auth = time.time()
 
 
 def main(
@@ -442,10 +439,10 @@ def main(
             Device to run inference on ("cpu" or "cuda")
     """
     # run kinit and htgettoken
-    # if server != "local":
-    #     logging.info("Authenticating")
-    #     authenticate()
-    #     logging.info("Authentication complete")
+    if server != "local":
+        logging.info("Authenticating")
+        authenticate()
+        logging.info("Authentication complete")
 
     fftlength = fftlength or kernel_length + fduration
     data = torch.randn(samples_per_event * len(inference_params))
@@ -527,27 +524,11 @@ def main(
     aframe = aframe.to(device)
 
     logging.info(f"Loading AMPLFI from weights at path {amplfi_weights}")
-    # amplfi, scaler = load_amplfi(
-    #     amplfi_architecture, amplfi_weights, len(inference_params)
-    # )
-    # amplfi = amplfi.to(device)
-    # scaler = scaler.to(device)
-    embedding_net = ResNet(
-        num_ifos=len(ifos),
-        layers=[5, 3, 3],
-        norm_layer=GroupNorm1DGetter(8),
-        context_dim=8,
+    amplfi, scaler = load_amplfi(
+        amplfi_architecture, amplfi_weights, len(inference_params)
     )
-    amplfi = InverseAutoregressiveFlow(
-        num_params=len(inference_params),
-        hidden_features=150,
-        num_transforms=80,
-        num_blocks=6,
-        # context_dim=8,
-        embedding_net=embedding_net,
-    ).to(device)
-    scaler = ChannelWiseScaler(len(inference_params)).to(device)
-    scaler.fit(torch.randn((8, 2048)))
+    amplfi = amplfi.to(device)
+    scaler = scaler.to(device)
 
     spectral_density = SpectralDensity(
         sample_rate=sample_rate,