diff --git a/zoobot/pytorch/estimators/define_model.py b/zoobot/pytorch/estimators/define_model.py
index 39926543..9d4fda9b 100755
--- a/zoobot/pytorch/estimators/define_model.py
+++ b/zoobot/pytorch/estimators/define_model.py
@@ -239,15 +239,11 @@ def configure_optimizers(self):
 
 
     def log_outputs(self, outputs, step_name):
-        self.log("{}/epoch_loss".format(step_name), outputs['loss'], on_epoch=True, on_step=False,prog_bar=True, logger=True, rank_zero_only=True)
-        # if self.log_on_step:
-        #     # seperate call to allow for different name, to allow for consistency with TF.keras auto-names
+        # self.log("{}/epoch_loss".format(step_name), outputs['loss'], on_epoch=True, on_step=False,prog_bar=True, logger=True, rank_zero_only=True)
+        # if outputs['predictions'].shape[1] == 2:  # will only do for binary classifications
         #     self.log(
-        #         "{}/step_loss".format(step_name), outputs['loss'], on_epoch=False, on_step=True, prog_bar=True, logger=True, rank_zero_only=True)
-        if outputs['predictions'].shape[1] == 2:  # will only do for binary classifications
-            # logging.info(predictions.shape, labels.shape)
-            self.log(
-                "{}_accuracy".format(step_name), self.train_accuracy(outputs['predictions'], torch.argmax(outputs['labels'], dim=1, keepdim=False)), prog_bar=True, rank_zero_only=True)
+        #         "{}_accuracy".format(step_name), self.train_accuracy(outputs['predictions'], torch.argmax(outputs['labels'], dim=1, keepdim=False)), prog_bar=True, rank_zero_only=True)
+        pass
 
 
     def log_loss_per_question(self, multiq_loss, prefix):
@@ -255,8 +251,9 @@ def log_loss_per_question(self, multiq_loss, prefix):
         # TODO need schema attribute or similar to have access to question names, this will do for now
         # unlike Finetuneable..., does not use TorchMetrics, simply logs directly
         # TODO could use TorchMetrics and for q in schema, self.q_metric loop
-        for question_n in range(multiq_loss.shape[1]):
-            self.log(f'{prefix}/epoch_questions/question_{question_n}_loss:0', torch.mean(multiq_loss[:, question_n]), on_epoch=True, on_step=False, rank_zero_only=True)
+        # for question_n in range(multiq_loss.shape[1]):
+        #     self.log(f'{prefix}/epoch_questions/question_{question_n}_loss:0', torch.mean(multiq_loss[:, question_n]), on_epoch=True, on_step=False, rank_zero_only=True)
+        pass
 
 
     
diff --git a/zoobot/pytorch/training/train_with_pytorch_lightning.py b/zoobot/pytorch/training/train_with_pytorch_lightning.py
index 8e8ad050..33e724a2 100644
--- a/zoobot/pytorch/training/train_with_pytorch_lightning.py
+++ b/zoobot/pytorch/training/train_with_pytorch_lightning.py
@@ -275,9 +275,9 @@ def train_default_zoobot_from_scratch(
             save_top_k=save_top_k
     )
 
-    early_stopping_callback = EarlyStopping(monitor='validation/epoch_loss', patience=patience, check_finite=True)
-
-    callbacks = [checkpoint_callback, early_stopping_callback] + extra_callbacks
+    # early_stopping_callback = EarlyStopping(monitor='validation/epoch_loss', patience=patience, check_finite=True)
+    # , early_stopping_callback
+    callbacks = [checkpoint_callback] + extra_callbacks
 
     trainer = pl.Trainer(
         log_every_n_steps=150,  # at batch 512 (A100 MP max), DR5 has ~161 train steps
@@ -290,12 +290,12 @@ def train_default_zoobot_from_scratch(
         callbacks=callbacks,
         max_epochs=epochs,
         default_root_dir=save_dir,
-        plugins=plugins,
-        use_distributed_sampler=use_distributed_sampler
+        plugins=plugins
+        # use_distributed_sampler=use_distributed_sampler
     )
 
-    logging.info((trainer.strategy, trainer.world_size,
-                 trainer.local_rank, trainer.global_rank, trainer.node_rank))
+    # logging.info((trainer.strategy, trainer.world_size,
+    #              trainer.local_rank, trainer.global_rank, trainer.node_rank))
 
     trainer.fit(lightning_model, datamodule)  # uses batch size of datamodule