_XFAIL_GLOO_WINDOWS

Author: Borda

src/lightning/fabric/utilities/imports.py

@@ -35,5 +35,5 @@
 _TORCH_GREATER_EQUAL_2_4 = compare_version("torch", operator.ge, "2.4.0")
 _TORCH_GREATER_EQUAL_2_4_1 = compare_version("torch", operator.ge, "2.4.1")
 _TORCH_LESS_EQUAL_2_6 = compare_version("torch", operator.le, "2.6.0")
-
+_TORCHMETRICS_GREATER_EQUAL_1_0_0 = compare_version("torchmetrics", operator.ge, "1.0.0")
 _PYTHON_GREATER_EQUAL_3_10_0 = (sys.version_info.major, sys.version_info.minor) >= (3, 10)

src/lightning/fabric/utilities/spike.py

@@ -1,19 +1,16 @@
 import json
-import operator
 import os
 import warnings
 from typing import TYPE_CHECKING, Any, Literal, Optional, Union
 
 import torch
-from lightning_utilities.core.imports import compare_version
 
+from lightning.fabric.utilities.imports import _TORCHMETRICS_GREATER_EQUAL_1_0_0
 from lightning.fabric.utilities.types import _PATH
 
 if TYPE_CHECKING:
     from lightning.fabric.fabric import Fabric
 
-_TORCHMETRICS_GREATER_EQUAL_1_0_0 = compare_version("torchmetrics", operator.ge, "1.0.0")
-
 
 class SpikeDetection:
     """Spike Detection Callback.

src/lightning/pytorch/trainer/connectors/logger_connector/result.py

@@ -25,9 +25,9 @@
 from lightning.fabric.utilities import move_data_to_device
 from lightning.fabric.utilities.apply_func import convert_tensors_to_scalars
 from lightning.fabric.utilities.distributed import _distributed_is_initialized
+from lightning.fabric.utilities.imports import _TORCHMETRICS_GREATER_EQUAL_1_0_0
 from lightning.pytorch.utilities.data import extract_batch_size
 from lightning.pytorch.utilities.exceptions import MisconfigurationException
-from lightning.pytorch.utilities.imports import _TORCHMETRICS_GREATER_EQUAL_1_0_0
 from lightning.pytorch.utilities.memory import recursive_detach
 from lightning.pytorch.utilities.rank_zero import WarningCache, rank_zero_warn
 from lightning.pytorch.utilities.warnings import PossibleUserWarning

src/lightning/pytorch/utilities/imports.py

@@ -25,6 +25,7 @@
 _TORCHMETRICS_GREATER_EQUAL_0_9_1 = RequirementCache("torchmetrics>=0.9.1")
 _TORCHMETRICS_GREATER_EQUAL_0_11 = RequirementCache("torchmetrics>=0.11.0")  # using new API with task
 _TORCHMETRICS_GREATER_EQUAL_1_0_0 = RequirementCache("torchmetrics>=1.0.0")
+_TORCH_GREATER_EQUAL_2_8 = RequirementCache("torch>=2.8.0")
 
 _OMEGACONF_AVAILABLE = package_available("omegaconf")
 _TORCHVISION_AVAILABLE = RequirementCache("torchvision")

tests/tests_fabric/utilities/test_spike.py

@@ -4,7 +4,8 @@
 import torch
 
 from lightning.fabric import Fabric
-from lightning.fabric.utilities.spike import _TORCHMETRICS_GREATER_EQUAL_1_0_0, SpikeDetection, TrainingSpikeException
+from lightning.fabric.utilities.imports import _TORCHMETRICS_GREATER_EQUAL_1_0_0
+from lightning.fabric.utilities.spike import SpikeDetection, TrainingSpikeException
 from tests_fabric.helpers.runif import RunIf
 
 

tests/tests_pytorch/callbacks/test_spike.py

@@ -3,9 +3,11 @@
 import pytest
 import torch
 
-from lightning.fabric.utilities.spike import _TORCHMETRICS_GREATER_EQUAL_1_0_0, TrainingSpikeException
+from lightning.fabric.utilities.imports import _IS_WINDOWS, _TORCHMETRICS_GREATER_EQUAL_1_0_0
+from lightning.fabric.utilities.spike import TrainingSpikeException
 from lightning.pytorch import LightningModule, Trainer
 from lightning.pytorch.callbacks.spike import SpikeDetection
+from lightning.pytorch.utilities.imports import _TORCH_GREATER_EQUAL_2_8
 from tests_pytorch.helpers.runif import RunIf
 
 
@@ -47,6 +49,10 @@ def on_train_batch_end(self, trainer, pl_module, outputs, batch, batch_idx):
             super().on_train_batch_end(trainer, pl_module, outputs, batch, batch_idx)
 
 
+# todo: RuntimeError: makeDeviceForHostname(): unsupported gloo device
+_XFAIL_GLOO_WINDOWS = pytest.mark.xfail(RuntimeError, strict=True, condition=(_IS_WINDOWS and _TORCH_GREATER_EQUAL_2_8))
+
+
 @pytest.mark.flaky(max_runs=3)
 @pytest.mark.parametrize(
     ("global_rank_spike", "num_devices", "spike_value", "finite_only"),
@@ -55,12 +61,12 @@ def on_train_batch_end(self, trainer, pl_module, outputs, batch, batch_idx):
     [
         pytest.param(0, 1, None, True),
         pytest.param(0, 1, None, False),
-        pytest.param(0, 1, float("inf"), True),
-        pytest.param(0, 1, float("inf"), False),
-        pytest.param(0, 1, float("-inf"), True),
-        pytest.param(0, 1, float("-inf"), False),
-        pytest.param(0, 1, float("NaN"), True),
-        pytest.param(0, 1, float("NaN"), False),
+        pytest.param(0, 1, float("inf"), True, marks=_XFAIL_GLOO_WINDOWS),
+        pytest.param(0, 1, float("inf"), False, marks=_XFAIL_GLOO_WINDOWS),
+        pytest.param(0, 1, float("-inf"), True, marks=_XFAIL_GLOO_WINDOWS),
+        pytest.param(0, 1, float("-inf"), False, marks=_XFAIL_GLOO_WINDOWS),
+        pytest.param(0, 1, float("NaN"), True, marks=_XFAIL_GLOO_WINDOWS),
+        pytest.param(0, 1, float("NaN"), False, marks=_XFAIL_GLOO_WINDOWS),
         pytest.param(0, 2, None, True, marks=RunIf(linux_only=True)),
         pytest.param(0, 2, None, False, marks=RunIf(linux_only=True)),
         pytest.param(1, 2, None, True, marks=RunIf(linux_only=True)),