Skip to content

Load transformer models saved on GPU on CPU-only machines #327

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
maxteneff wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Load transformer models saved on GPU on CPU-only machines #327

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
17b5961
Load transformer models saved on GPU on CPU-only machines
maxteneff Oct 17, 2026
2117a2b
Link the changelog entry to the pull request
maxteneff Oct 17, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 5 additions & 0 deletions CHANGELOG.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,6 +6,11 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).


## Unreleased

### Fixed
- Loading fitted transformer models (`SASRecModel`, `BERT4RecModel`) that were saved on GPU with `load_model`, `load` and `loads` on a machine without GPU ([#327](https://github.com/MTSWebServices/RecTools/pull/327))

## [0.19.0] - 11.06.2026

### Added
Expand Down
4 changes: 3 additions & 1 deletion rectools/models/nn/transformers/base.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -668,7 +668,9 @@ def __getstate__(self) -> object:

def __setstate__(self, state: tp.Dict[str, tp.Any]) -> None:
if "fitted_checkpoint" in state:
checkpoint = torch.load(io.BytesIO(state["fitted_checkpoint"]), weights_only=False)
# Load to CPU so that a model fitted on GPU can be loaded on a machine without one.
# Weights are moved to the right device before inference and by the trainer when fitting.
checkpoint = torch.load(io.BytesIO(state["fitted_checkpoint"]), map_location="cpu", weights_only=False)
loaded = self._model_from_checkpoint(checkpoint)
else:
loaded = self.from_config(state["model_config"])
Expand Down
20 changes: 20 additions & 0 deletions tests/models/nn/transformers/test_base.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -184,6 +184,26 @@ def test_save_load_for_fitted_model(
model.fit(dataset_item_features)
assert_save_load_do_not_change_model(model, dataset_item_features)

@pytest.mark.parametrize("model_cls", (SASRecModel, BERT4RecModel))
def test_load_fitted_model_saved_on_gpu_without_gpu(
self,
model_cls: tp.Type[TransformerModelBase],
dataset: Dataset,
monkeypatch: pytest.MonkeyPatch,
) -> None:
model = model_cls.from_config({"deterministic": True})
model.fit(dataset)

# Save the model as if its tensors were on a GPU
with monkeypatch.context() as m:
m.setattr(torch.serialization, "location_tag", lambda storage: "cuda:0")
data = model.dumps()

monkeypatch.setattr(torch.cuda, "is_available", lambda: False)
recovered_model = model_cls.loads(data)

self._assert_same_reco(model, recovered_model, dataset)

@pytest.mark.parametrize("test_dataset", ("dataset", "dataset_item_features"))
@pytest.mark.parametrize("model_cls", (SASRecModel, BERT4RecModel))
@pytest.mark.parametrize(
Expand Down