Apply suggestions from code review

accept suggestions

Co-authored-by: Daria <93913290+blondered@users.noreply.github.com>

Author: nsundalov

CHANGELOG.md

@@ -8,7 +8,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## Unreleased
 ### Added
 - `extras` argument to `SequenceDataset`, `extra_cols` argument to `TransformerDataPreparatorBase`, `session_tower_forward` and `item_tower_forward` methods to `SimilarityModuleBase` ([#287](https://github.com/MobileTeleSystems/RecTools/pull/287))
-- allow resaving transformer model multiple times. Load train state on model loading ([#289](https://github.com/MobileTeleSystems/RecTools/pull/289))
+- Support for resaving transformer models multiple times and loading trainer state ([#289](https://github.com/MobileTeleSystems/RecTools/pull/289))
 
 ## [0.14.0] - 16.05.2025
 

rectools/models/nn/transformers/base.py

@@ -610,8 +610,7 @@ def _model_from_checkpoint(cls, checkpoint: tp.Dict[str, tp.Any]) -> tpe.Self:
         # save checkpoint to temp file to be able to use it in trainer
         with NamedTemporaryFile() as f:
             torch.save(checkpoint, f.name)
-            fit_trainer = deepcopy(loaded._trainer)
-            loaded.fit_trainer = fit_trainer
+            loaded.fit_trainer = deepcopy(loaded._trainer)
             # use stub dataset to load trainer state
             loaded.fit_trainer.fit(
                 loaded.lightning_model,