fix: issue #13

shscripts/train_cogvideox_lora.sh

@@ -14,4 +14,4 @@ python scripts/train.py \
 --logdir $RESROOT \
 --devices '0,' \
 lightning.trainer.num_nodes=1 \
---auto_resume False
+--auto_resume

shscripts/train_dynamicrafter.sh

@@ -21,5 +21,5 @@ python scripts/train.py \
 --ckpt $DCCKPT \
 --devices '0,' \
 lightning.trainer.num_nodes=1 \
---auto_resume True
+--auto_resume
 

shscripts/train_opensorav10.sh

@@ -13,4 +13,4 @@ lightning.trainer.num_nodes=1 \
 --base $CONFIG \
 --name "$current_time"_$EXPNAME \
 --logdir $LOGDIR \
---auto_resume True
+--auto_resume

shscripts/train_videocrafter_v2.sh

@@ -20,4 +20,4 @@ python scripts/train.py \
 --name ${CURRENT_TIME}_${EXPNAME} \
 --devices '0,' \
 lightning.trainer.num_nodes=1 \
---auto_resume True
+--auto_resume

src/base/ddpm3d.py

@@ -601,7 +601,6 @@ def differentiable_decode_first_stage(self, z, **kwargs):
         """same as decode_first_stage but without decorator"""
         return self._decode_core(z, **kwargs)
 
-    @torch.no_grad()
     def get_batch_input(self, batch, random_uncond, return_first_stage_outputs=False, return_original_cond=False, is_imgbatch=False):
         ## image/video shape: b, c, t, h, w
         data_key = 'jpg' if is_imgbatch else self.first_stage_key

src/utils/train_utils.py

@@ -270,12 +270,19 @@ def get_autoresume_path(logdir):
 def set_logger(logfile, name='mainlogger'):
     logger = logging.getLogger(name)
     logger.setLevel(logging.INFO)
+
+    # Set the logger to prevent log propagation to the parent logger and print twice.
+    logger.propagate = False
+
     fh = logging.FileHandler(logfile, mode='w')
     fh.setLevel(logging.INFO)
+
     ch = logging.StreamHandler()
     ch.setLevel(logging.DEBUG)
+
     fh.setFormatter(logging.Formatter("%(asctime)s-%(levelname)s: %(message)s"))
     ch.setFormatter(logging.Formatter("%(message)s"))
+
     logger.addHandler(fh)
     logger.addHandler(ch)
     return logger