Fix batch size extraction when set by the user in `LightningModule.lo…

…g` (Lightning-AI#10408)

Co-authored-by: Carlos Mocholi <carlossmocholi@gmail.com>

CHANGELOG.md

@@ -158,6 +158,9 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - Fixed signals being registered within threads ([#10610](https://github.com/PyTorchLightning/pytorch-lightning/pull/10610))
 
 
+- Fixed an issue that caused Lightning to extract the batch size even though it was set by the user in `LightningModule.log` ([#10408](https://github.com/PyTorchLightning/pytorch-lightning/pull/10408))
+
+
 - Fixed uploading best model checkpoint in NeptuneLogger ([#10369](https://github.com/PyTorchLightning/pytorch-lightning/pull/10369))
 
 

pytorch_lightning/loops/epoch/training_epoch_loop.py

@@ -161,9 +161,7 @@ def advance(self, *args: Any, **kwargs: Any) -> None:
 
         self.batch_progress.increment_ready()
 
-        # cache the batch size value to avoid extracting it again after the batch loop runs as the value will be
-        # different if tbptt is enabled
-        batch_size = self.trainer.logger_connector.on_batch_start(batch_idx, batch)
+        self.trainer.logger_connector.on_batch_start(batch_idx, batch)
 
         if batch is None:
             self._warning_cache.warn("train_dataloader yielded None. If this was on purpose, ignore this warning...")
@@ -194,8 +192,6 @@ def advance(self, *args: Any, **kwargs: Any) -> None:
             with self.trainer.profiler.profile("run_training_batch"):
                 batch_output = self.batch_loop.run(batch, batch_idx)
 
-        self.trainer._results.batch_size = batch_size
-
         self.batch_progress.increment_processed()
 
         # update non-plateau LR schedulers

pytorch_lightning/trainer/connectors/logger_connector/logger_connector.py

@@ -214,7 +214,6 @@ def update_eval_epoch_metrics(self) -> List[_OUT_DICT]:
 
     def on_train_split_start(self, split_idx: int, split_batch: Any) -> None:
         self._split_idx = split_idx
-        self.on_new_batch(split_batch)
 
     def update_train_step_metrics(self) -> None:
         if self.trainer.fit_loop._should_accumulate() and self.trainer.lightning_module.automatic_optimization:
@@ -257,28 +256,23 @@ def _log_gpus_metrics(self) -> None:
     Utilities and properties
     """
 
-    def on_new_batch(self, batch: Any) -> int:
-        # when the user requests `dataloader_iter`, we can't track the batch_size
-        # and this is left to user responsibility.
-        if not isinstance(batch, pl.utilities.fetching.StepFuncDataLoaderIter):
-            assert self.trainer._results is not None
-            return self.trainer._results.extract_batch_size(batch)
-        return 1
-
     def on_epoch_start(self) -> None:
         self._epoch_end_reached = False
 
-    def on_batch_start(self, batch_idx: int, batch: Any) -> int:
+    def on_batch_start(self, batch_idx: int, batch: Any) -> None:
         self._batch_idx = batch_idx
         self._epoch_end_reached = False
-        return self.on_new_batch(batch)
+
+        assert self.trainer._results is not None
+        # attach reference to the new batch and remove the cached batch_size
+        self.trainer._results.batch = batch
+        self.trainer._results.batch_size = None
 
     def epoch_end_reached(self) -> None:
         self._epoch_end_reached = True
         self._batch_idx = None
         self._split_idx = None
         assert self.trainer._results is not None
-        self.trainer._results.batch_size = 1
 
     def on_epoch_end(self) -> None:
         assert self._epoch_end_reached
@@ -295,6 +289,11 @@ def on_batch_end(self) -> None:
         self._callback_metrics.update(metrics["callback"])
         self._logged_metrics.update(metrics["log"])
 
+        assert self.trainer._results is not None
+        # drop the reference to current batch and batch_size
+        self.trainer._results.batch = None
+        self.trainer._results.batch_size = None
+
     def should_reset_tensors(self, fx: str) -> bool:
         is_different_fx = self._current_fx != fx
         if self._split_idx is None:

pytorch_lightning/trainer/connectors/logger_connector/result.py

@@ -212,7 +212,7 @@ def __init__(self, metadata: _Metadata, is_tensor: bool) -> None:
             if self.meta.is_mean_reduction:
                 self.add_state("cumulated_batch_size", torch.tensor(0), dist_reduce_fx=torch.sum)
 
-    def update(self, value: _IN_METRIC, batch_size: torch.Tensor) -> None:
+    def update(self, value: _IN_METRIC, batch_size: int) -> None:
         if self.is_tensor:
             if not torch.is_floating_point(value):
                 dtype = torch.get_default_dtype()
@@ -259,7 +259,7 @@ def reset(self) -> None:
             self.value.reset()
         self.has_reset = True
 
-    def forward(self, value: _IN_METRIC, batch_size: torch.Tensor) -> None:
+    def forward(self, value: _IN_METRIC, batch_size: int) -> None:
         if self.meta.enable_graph:
             with torch.no_grad():
                 self.update(value, batch_size)
@@ -385,8 +385,9 @@ class ResultCollection(dict):
     def __init__(self, training: bool, device: Optional[Union[str, torch.device]] = None) -> None:
         super().__init__()
         self.training = training
-        self._batch_size = torch.tensor(1, device=device)
         self.device: Optional[Union[str, torch.device]] = device
+        self.batch: Optional[Any] = None
+        self.batch_size: Optional[int] = None
 
     @property
     def result_metrics(self) -> List[ResultMetric]:
@@ -399,14 +400,23 @@ def append_fn(v: ResultMetric) -> None:
         apply_to_collection(list(self.values()), ResultMetric, append_fn)
         return o
 
-    @property
-    def batch_size(self) -> torch.Tensor:
-        # performance: cache the `batch_size` tensor instead of re-creating it
-        return self._batch_size
+    def _extract_batch_size(self, batch_size: Optional[int], meta: _Metadata) -> int:
+        # check if we have extracted the batch size already
+        if batch_size is None:
+            batch_size = self.batch_size
+
+        if batch_size is not None:
+            return batch_size
 
-    @batch_size.setter
-    def batch_size(self, value: int) -> None:
-        self._batch_size = torch.tensor(value, device=self.device)
+        batch_size = 1
+        if self.batch is not None and meta.on_epoch and meta.is_mean_reduction:
+            try:
+                batch_size = extract_batch_size(self.batch)
+                self.batch_size = batch_size
+            except RecursionError:
+                pass
+
+        return batch_size
 
     def log(
         self,
@@ -467,10 +477,8 @@ def log(
                 f"You called `self.log({name}, ...)` twice in `{fx}` with different arguments. This is not allowed"
             )
 
-        if batch_size is not None:
-            self.batch_size = batch_size
-
-        self.update_metrics(key, value)
+        batch_size = self._extract_batch_size(batch_size, meta)
+        self.update_metrics(key, value, batch_size)
 
     def register_key(self, key: str, meta: _Metadata, value: _METRIC_COLLECTION) -> None:
         """Create one ResultMetric object per value.
@@ -487,10 +495,10 @@ def fn(v: _IN_METRIC) -> ResultMetric:
             value = ResultMetricCollection(value)
         self[key] = value
 
-    def update_metrics(self, key: str, value: _METRIC_COLLECTION) -> None:
-        def fn(result_metric: ResultMetric, v: ResultMetric) -> None:
+    def update_metrics(self, key: str, value: _METRIC_COLLECTION, batch_size: int) -> None:
+        def fn(result_metric: ResultMetric, v: torch.Tensor) -> None:
             # performance: avoid calling `__call__` to avoid the checks in `torch.nn.Module._call_impl`
-            result_metric.forward(v.to(self.device), self.batch_size)
+            result_metric.forward(v.to(self.device), batch_size)
             result_metric.has_reset = False
 
         apply_to_collections(self[key], value, ResultMetric, fn)
@@ -584,19 +592,10 @@ def fn(item: ResultMetric) -> None:
 
         apply_to_collection(self, ResultMetric, fn)
 
-    def extract_batch_size(self, batch: Any) -> int:
-        try:
-            batch_size = extract_batch_size(batch)
-        except RecursionError:
-            batch_size = 1
-        self.batch_size = batch_size  # the setter converts it to `Tensor`
-        return batch_size
-
     def to(self, *args: Any, **kwargs: Any) -> "ResultCollection":
         """Move all data to the given device."""
         self.update(apply_to_collection(dict(self), (torch.Tensor, Metric), move_data_to_device, *args, **kwargs))
 
-        self._batch_size = self._batch_size.to(*args, **kwargs)
         if "device" in kwargs:
             self.device = kwargs["device"]
         return self

pytorch_lightning/utilities/data.py

@@ -36,7 +36,10 @@
 
 def _extract_batch_size(batch: BType) -> Generator[int, None, None]:
     if isinstance(batch, torch.Tensor):
-        yield batch.size(0)
+        if batch.ndim == 0:
+            yield 1
+        else:
+            yield batch.size(0)
     elif isinstance(batch, str):
         yield len(batch)
     elif isinstance(batch, (Iterable, Mapping)):

tests/deprecated_api/__init__.py

@@ -14,7 +14,7 @@
 """Test deprecated functionality which will be removed in vX.Y.Z."""
 import sys
 from contextlib import contextmanager
-from typing import Optional
+from typing import Optional, Type
 
 import pytest
 
@@ -26,14 +26,28 @@ def _soft_unimport_module(str_module):
 
 
 @contextmanager
-def no_deprecated_call(match: Optional[str] = None):
+def no_warning_call(expected_warning: Type[Warning] = UserWarning, match: Optional[str] = None):
     with pytest.warns(None) as record:
         yield
+
+    if match is None:
         try:
-            w = record.pop(DeprecationWarning)
-            if match is not None and match not in str(w.message):
-                return
+            w = record.pop(expected_warning)
         except AssertionError:
-            # no DeprecationWarning raised
+            # no warning raised
+            return
+    else:
+        for w in record.list:
+            if w.category is expected_warning and match in w.message.args[0]:
+                break
+        else:
             return
-        raise AssertionError(f"`DeprecationWarning` was raised: {w}")
+
+    msg = "A warning" if expected_warning is None else f"`{expected_warning.__name__}`"
+    raise AssertionError(f"{msg} was raised: {w}")
+
+
+@contextmanager
+def no_deprecated_call(match: Optional[str] = None):
+    with no_warning_call(expected_warning=DeprecationWarning, match=match):
+        yield

tests/loops/test_loop_state_dict.py

@@ -14,7 +14,6 @@
 from unittest.mock import Mock
 
 import pytest
-import torch
 
 from pytorch_lightning.loops import FitLoop
 from pytorch_lightning.trainer.trainer import Trainer
@@ -80,14 +79,16 @@ def test_loops_state_dict_structure():
                 "is_last_batch": False,
             },
             "epoch_loop.val_loop._results": {
+                "batch": None,
+                "batch_size": None,
                 "training": False,
-                "_batch_size": torch.tensor(1),
                 "device": None,
                 "items": {},
             },
             "epoch_loop._results": {
+                "batch": None,
+                "batch_size": None,
                 "training": True,
-                "_batch_size": torch.tensor(1),
                 "device": None,
                 "items": {},
             },
@@ -106,8 +107,9 @@ def test_loops_state_dict_structure():
                 "is_last_batch": False,
             },
             "_results": {
+                "batch": None,
+                "batch_size": None,
                 "training": False,
-                "_batch_size": torch.tensor(1),
                 "device": None,
                 "items": {},
             },
@@ -122,8 +124,9 @@ def test_loops_state_dict_structure():
                 "is_last_batch": False,
             },
             "_results": {
+                "batch": None,
+                "batch_size": None,
                 "training": False,
-                "_batch_size": torch.tensor(1),
                 "device": None,
                 "items": {},
             },

tests/trainer/logging_/test_train_loop_logging.py

@@ -27,6 +27,7 @@
 from pytorch_lightning.callbacks import EarlyStopping, ModelCheckpoint, TQDMProgressBar
 from pytorch_lightning.core.lightning import LightningModule
 from pytorch_lightning.utilities.exceptions import MisconfigurationException
+from tests.deprecated_api import no_warning_call
 from tests.helpers.boring_model import BoringModel, RandomDataset, RandomDictDataset
 from tests.helpers.runif import RunIf
 
@@ -715,19 +716,15 @@ def on_validation_epoch_end(self):
             assert all(v == 3 for v in self.trainer.callback_metrics.values())
 
         def on_train_batch_start(self, batch, batch_idx):
-            assert self.trainer._results.batch_size == 2
             self.log("on_train_batch_start", 1.0, reduce_fx="sum")
 
         def on_train_batch_end(self, outputs, batch, batch_idx):
-            assert self.trainer._results.batch_size == 2
             self.log("on_train_batch_end", 1.0, reduce_fx="sum")
 
         def on_validation_batch_start(self, batch, batch_idx, dataloader_idx):
-            assert self.trainer._results.batch_size == 2
             self.log("on_validation_batch_start", 1.0, reduce_fx="sum")
 
         def on_validation_batch_end(self, outputs, batch, batch_idx, dataloader_idx):
-            assert self.trainer._results.batch_size == 2
             self.log("on_validation_batch_end", 1.0, reduce_fx="sum")
 
         def training_epoch_end(self, *_) -> None:
@@ -749,3 +746,36 @@ def validation_epoch_end(self, *_) -> None:
     train_data = DataLoader(RandomDataset(32, 64), batch_size=2)
     val_data = DataLoader(RandomDataset(32, 64), batch_size=2)
     trainer.fit(model, train_dataloaders=train_data, val_dataloaders=val_data)
+
+
+def test_no_batch_size_extraction_with_specifying_explictly(tmpdir):
+    batch_size = BoringModel().train_dataloader().batch_size + 1
+    fast_dev_run = 2
+    log_val = 7
+
+    class CustomBoringModel(BoringModel):
+        def on_before_batch_transfer(self, batch, *args, **kwargs):
+            # This is an ambiguous batch which have multiple potential batch sizes
+            if self.trainer.training:
+                batch = {"batch1": torch.randn(batch_size, 10), "batch2": batch}
+            return batch
+
+        def training_step(self, batch, batch_idx):
+            self.log("step_log_val", log_val, on_epoch=False)
+            self.log("epoch_log_val", log_val, batch_size=batch_size, on_step=False, on_epoch=True)
+            self.log("epoch_sum_log_val", log_val, on_epoch=True, reduce_fx="sum")
+            return super().training_step(batch["batch2"], batch_idx)
+
+        def on_train_epoch_end(self, *args, **kwargs):
+            results = self.trainer._results
+            assert results["training_step.step_log_val"].value == log_val
+            assert results["training_step.step_log_val"].cumulated_batch_size == 0
+            assert results["training_step.epoch_log_val"].value == log_val * batch_size * fast_dev_run
+            assert results["training_step.epoch_log_val"].cumulated_batch_size == batch_size * fast_dev_run
+            assert results["training_step.epoch_sum_log_val"].value == log_val * fast_dev_run
+
+    model = CustomBoringModel()
+    trainer = Trainer(default_root_dir=tmpdir, fast_dev_run=fast_dev_run)
+
+    with no_warning_call(match="Trying to infer the `batch_size`"):
+        trainer.fit(model)

tests/utilities/test_data.py

@@ -12,16 +12,16 @@
     warning_cache,
 )
 from pytorch_lightning.utilities.exceptions import MisconfigurationException
+from tests.deprecated_api import no_warning_call
 from tests.helpers.boring_model import BoringModel, RandomDataset, RandomIterableDataset
 
 
 def test_extract_batch_size():
     """Tests the behavior of extracting the batch size."""
 
     def _check_warning_not_raised(data, expected):
-        with pytest.warns(None) as record:
+        with no_warning_call(match="Trying to infer the `batch_size`"):
             assert extract_batch_size(data) == expected
-        assert len(record) == 0
 
     def _check_warning_raised(data, expected):
         with pytest.warns(UserWarning, match=f"Trying to infer the `batch_size` .* we found is {expected}."):
@@ -43,6 +43,9 @@ def _check_warning_raised(data, expected):
     batch = {"test": [{"test": [torch.zeros(11, 10)]}]}
     _check_warning_not_raised(batch, 11)
 
+    batch = {"a": [torch.tensor(1), torch.tensor(2)], "b": torch.tensor([1, 2, 3, 4])}
+    _check_warning_raised(batch, 1)
+
     batch = {"test": [{"test": [torch.zeros(11, 10), torch.zeros(10, 10)]}]}
     _check_warning_raised(batch, 11)