Fix moving keys to device in ResultCollection

src/lightning/pytorch/CHANGELOG.md

@@ -15,6 +15,7 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 
 - Added a flag `verbose` to the `seed_everything()` function ([#20108](https://github.com/Lightning-AI/pytorch-lightning/pull/20108))
 
+-
 
 ### Changed
 
@@ -44,6 +45,8 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 
 - Fixed an issue that would cause too many printouts of the seed info when using `seed_everything()` ([#20108](https://github.com/Lightning-AI/pytorch-lightning/pull/20108))
 
+- Fixed `_LoggerConnector`'s `_ResultMetric` to move all registered keys to the device of the logged value if needed ([#19814](https://github.com/Lightning-AI/pytorch-lightning/issues/19814))
+
 
 
 ## [2.3.0] - 2024-06-13

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

@@ -400,26 +400,19 @@ def log(
 
         # register logged value if it doesn't exist
         if key not in self:
-            self.register_key(key, meta, value)
+            metric = _ResultMetric(meta, isinstance(value, Tensor))
+            self[key] = metric
 
         # check the stored metadata and the current one match
         elif meta != self[key].meta:
             raise MisconfigurationException(
                 f"You called `self.log({name}, ...)` twice in `{fx}` with different arguments. This is not allowed"
             )
+        self[key].to(value.device)
 
         batch_size = self._extract_batch_size(self[key], batch_size, meta)
         self.update_metrics(key, value, batch_size)
 
-    def register_key(self, key: str, meta: _Metadata, value: _VALUE) -> None:
-        """Create one _ResultMetric object per value.
-
-        Value can be provided as a nested collection
-
-        """
-        metric = _ResultMetric(meta, isinstance(value, Tensor)).to(value.device)
-        self[key] = metric
-
     def update_metrics(self, key: str, value: _VALUE, batch_size: int) -> None:
         result_metric = self[key]
         # performance: avoid calling `__call__` to avoid the checks in `torch.nn.Module._call_impl`

tests/tests_pytorch/trainer/logging_/test_logger_connector.py

@@ -33,6 +33,7 @@
 from torchmetrics import Accuracy, MeanAbsoluteError, MeanSquaredError, MetricCollection
 from torchmetrics import AveragePrecision as AvgPre
 
+from tests_pytorch.helpers.runif import RunIf
 from tests_pytorch.models.test_hooks import get_members
 
 
@@ -639,3 +640,23 @@ def test_result_collection_no_batch_size_extraction():
     assert results["training_step.epoch_log_val"].value == log_val * batch_size
     assert results["training_step.epoch_log_val"].cumulated_batch_size == batch_size
     assert results["training_step.epoch_sum_log_val"].value == log_val
+
+
+@RunIf(min_cuda_gpus=1)
+def test_result_collection_changes_device():
+    """Test that the keys in the ResultCollection are moved to the device together with the collection."""
+    results = _ResultCollection(training=True)
+    fx, name = "training_step", "step_log_val"
+    log_val = torch.tensor(7.0, device="cuda:0")
+
+    # same device as the original tensor
+    results.log(fx, name, log_val, on_step=True, on_epoch=False, reduce_fx="mean")
+    assert results[f"{fx}.{name}"].cumulated_batch_size.device == log_val.device
+
+    # moved to cpu
+    results.cpu()
+    assert results[f"{fx}.{name}"].cumulated_batch_size.device == torch.device("cpu")
+
+    # same device as the new tensor
+    results.log(fx, name, log_val, on_step=True, on_epoch=False, reduce_fx="mean")
+    assert results[f"{fx}.{name}"].cumulated_batch_size.device == log_val.device