A reactive notebook for Python — run reproducible experiments, query with SQL, execute as a script, deploy as an app, and version with git. Stored as pure Python. All in a modern, AI-native editor.
Python artificial-intelligencedagdata-sciencedata-visualizationdataflowdeveloper-toolsmachine-learningnotebookspipelinepythonreactivesqlweb-app
# Copyright 2026 Marimo. All rights reserved.
from __future__ import annotations
import pytest
torch = pytest.importorskip("torch")
from marimo._save.loaders.lazy import maybe_update_lazy_stub # noqa: E402
from marimo._save.stubs.lazy_stub import ( # noqa: E402
BLOB_DESERIALIZERS,
BLOB_SERIALIZERS,
)
def test_tensor_resolves_to_pt_codec() -> None:
assert maybe_update_lazy_stub(torch.ones(4)) == "pt"
def test_parameter_resolves_to_pt_codec_via_mro() -> None:
param = torch.nn.Parameter(torch.ones(4))
assert maybe_update_lazy_stub(param) == "pt"
def test_pt_round_trip() -> None:
tensor = torch.randn(8, 3)
data = BLOB_SERIALIZERS["pt"](tensor)
restored = BLOB_DESERIALIZERS[".pt"](data, "torch.Tensor")
assert torch.equal(tensor, restored)
assert restored.dtype == tensor.dtype
def test_pt_round_trip_preserves_parameter_subclass() -> None:
param = torch.nn.Parameter(torch.randn(4))
data = BLOB_SERIALIZERS["pt"](param)
restored = BLOB_DESERIALIZERS[".pt"](data, "torch.nn.parameter.Parameter")
assert isinstance(restored, torch.nn.Parameter)
assert torch.equal(param.data, restored.data)