fix: Fix LRSchedulerWrapper with shared memory address issue.

vis4d/config/default/pl_trainer.py

@@ -1,7 +1,7 @@
 """Default runtime configuration for PyTorch Lightning."""
 import inspect
 
-import pytorch_lightning as pl
+from lightning import Trainer
 
 from vis4d.config import FieldConfigDict
 from vis4d.config.typing import ExperimentConfig
@@ -12,7 +12,7 @@ def get_default_pl_trainer_cfg(config: ExperimentConfig) -> ExperimentConfig:
     pl_trainer = FieldConfigDict()
 
     # PL Trainer arguments
-    for k, v in inspect.signature(pl.Trainer).parameters.items():
+    for k, v in inspect.signature(Trainer).parameters.items():
         if not k in {"callbacks", "devices", "logger", "strategy"}:
             pl_trainer[k] = v.default
 

vis4d/engine/callbacks/logging.py

@@ -51,6 +51,7 @@ def on_train_epoch_start(
         """Hook to run at the start of a training epoch."""
         if self.epoch_based:
             self.train_timer.reset()
+            self.last_step = 0
             self._metrics.clear()
         elif trainer_state["global_step"] == 0:
             self.train_timer.reset()

vis4d/engine/optim/scheduler.py

@@ -7,7 +7,7 @@
 from torch.optim.lr_scheduler import LRScheduler
 
 from vis4d.common.typing import DictStrAny
-from vis4d.config import instantiate_classes
+from vis4d.config import copy_and_resolve_references, instantiate_classes
 from vis4d.config.typing import LrSchedulerConfig
 
 
@@ -30,7 +30,9 @@ def __init__(
         steps_per_epoch: int = -1,
     ) -> None:
         """Initialize LRSchedulerWrapper."""
-        self.lr_schedulers_cfg = lr_schedulers_cfg
+        self.lr_schedulers_cfg: list[
+            LrSchedulerConfig
+        ] = copy_and_resolve_references(lr_schedulers_cfg)
         self.lr_schedulers: dict[int, LRSchedulerDict] = {}
         super().__init__(optimizer)
         self.steps_per_epoch = steps_per_epoch