Move torchmetrics to device when using FSDP (#18954)

src/lightning/fabric/strategies/fsdp.py

@@ -33,6 +33,7 @@
 )
 
 import torch
+from lightning_utilities.core.imports import RequirementCache
 from torch import Tensor
 from torch.nn import Module, Parameter
 from torch.optim import Optimizer
@@ -292,6 +293,8 @@ def setup_module(self, module: Module) -> Module:
                 **self._fsdp_kwargs,
             )
 
+        _move_torchmetrics_to_device(module, self.root_device)
+
         # activation checkpointing needs to be set up after wrapping the model
         if _TORCH_GREATER_EQUAL_1_13:
             _setup_activation_checkpointing(module, self._activation_checkpointing_kwargs)
@@ -886,3 +889,15 @@ def _has_meta_device_parameters(obj: Union[Module, Optimizer]) -> bool:
     if isinstance(obj, Module):
         return any(t.is_meta for t in obj.parameters())
     raise TypeError(f"Expected `torch.nn.Module` or `torch.optim.Optimizer`, got: {type(obj).__name__}")
+
+
+def _move_torchmetrics_to_device(module: torch.nn.Module, device: torch.device) -> None:
+    # FSDP doesn't move modules without parameters (e.g. Metrics) to the device
+    # https://github.com/pytorch/pytorch/issues/113113
+    if not RequirementCache("torchmetrics"):
+        return
+
+    from torchmetrics import Metric
+
+    for metric in (m for m in module.modules() if isinstance(m, Metric)):
+        metric.to(device)  # `.to()` is in-place

src/lightning/pytorch/CHANGELOG.md

@@ -49,6 +49,9 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - Fixed an issue causing permission errors on Windows when attempting to create a symlink for the "last" checkpoint ([#18942](https://github.com/Lightning-AI/lightning/issues/18942))
 
 
+- Fixed an issue where Metric instances from `torchmetrics` wouldn't get moved to the device when using FSDP ([#18954](https://github.com/Lightning-AI/lightning/issues/18954))
+
+
 ## [2.1.0] - 2023-10-11
 
 ### Added

src/lightning/pytorch/strategies/fsdp.py

@@ -39,6 +39,7 @@
     _is_full_checkpoint,
     _is_sharded_checkpoint,
     _load_raw_module_state,
+    _move_torchmetrics_to_device,
     _optimizer_has_flat_params,
     _setup_activation_checkpointing,
 )
@@ -292,6 +293,8 @@ def _setup_model(self, model: Module) -> Module:
                 **self.kwargs,
             )
 
+        _move_torchmetrics_to_device(model, self.root_device)
+
         # activation checkpointing needs to be set up after wrapping the model
         if _TORCH_GREATER_EQUAL_1_13:
             _setup_activation_checkpointing(model, self._activation_checkpointing_kwargs)

tests/tests_pytorch/strategies/test_fsdp.py

@@ -28,6 +28,7 @@
 from lightning.pytorch.utilities.exceptions import MisconfigurationException
 from torch.distributed.fsdp.fully_sharded_data_parallel import CPUOffload, FullyShardedDataParallel, MixedPrecision
 from torch.distributed.fsdp.wrap import always_wrap_policy, size_based_auto_wrap_policy, wrap
+from torchmetrics import Accuracy
 
 from tests_pytorch.helpers.runif import RunIf
 
@@ -239,6 +240,36 @@ def test_fsdp_strategy_sync_batchnorm(tmpdir):
     _run_multiple_stages(trainer, model, os.path.join(tmpdir, "last.ckpt"))
 
 
+@RunIf(min_cuda_gpus=1, skip_windows=True)
+def test_fsdp_modules_without_parameters(tmp_path):
+    """Test that TorchMetrics get moved to the device despite not having any parameters."""
+
+    class MetricsModel(BoringModel):
+        def __init__(self):
+            super().__init__()
+            self.metric = Accuracy("multiclass", num_classes=10)
+            assert self.metric.device == self.metric.tp.device == torch.device("cpu")
+
+        def setup(self, stage) -> None:
+            assert self.metric.device == self.metric.tp.device == torch.device("cpu")
+
+        def training_step(self, batch, batch_idx):
+            loss = super().training_step(batch, batch_idx)
+            assert self.metric.device == self.metric.tp.device == torch.device("cuda", 0)
+            self.metric(torch.rand(2, 10, device=self.device), torch.randint(0, 10, size=(2,), device=self.device))
+            return loss
+
+    model = MetricsModel()
+    trainer = Trainer(
+        default_root_dir=tmp_path,
+        accelerator="cuda",
+        devices=1,
+        strategy="fsdp",
+        max_steps=1,
+    )
+    trainer.fit(model)
+
+
 @RunIf(min_cuda_gpus=1, skip_windows=True, standalone=True)
 @pytest.mark.parametrize("precision", ["16-mixed", pytest.param("bf16-mixed", marks=RunIf(bf16_cuda=True))])
 def test_fsdp_strategy_checkpoint(tmpdir, precision):