Merge branch 'master' into refactor/remove-check-ckpt-callback

Author: carmocca

CHANGELOG.md

@@ -166,6 +166,9 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - Added support for `accelerator='cpu'|'gpu'|'tpu'|'ipu'|'auto'` ([#7808](https://github.com/PyTorchLightning/pytorch-lightning/pull/7808))
 
 
+- Added `tpu_spawn_debug` to plugin registry ([#7933](https://github.com/PyTorchLightning/pytorch-lightning/pull/7933))
+
+
 - Enabled traditional/manual launching of DDP processes through `LOCAL_RANK` and `NODE_RANK` environment variable assignments ([#7480](https://github.com/PyTorchLightning/pytorch-lightning/pull/7480))
 
 
@@ -303,6 +306,12 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - `Trainer(resume_from_checkpoint=...)` now restores the model directly after `LightningModule.setup()`, which is before `LightningModule.configure_sharded_model()` ([#7652](https://github.com/PyTorchLightning/pytorch-lightning/pull/7652))
 
 
+- Moved `torch.cuda.set_device()` to enable collective calls earlier in setup ([#8312](https://github.com/PyTorchLightning/pytorch-lightning/pull/8312))
+
+
+- Use XLA utility API to move data to CPU (Single TPU core) ([#8078](https://github.com/PyTorchLightning/pytorch-lightning/pull/8078))
+
+
 ### Deprecated
 
 
@@ -460,6 +469,9 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - Fixed missing call to `LightningModule.untoggle_optimizer` in training loop when running gradient accumulation with multiple optimizers ([#8284](https://github.com/PyTorchLightning/pytorch-lightning/pull/8284))
 
 
+- Fixed progress bar updates for Pod Training ([#8258](https://github.com/PyTorchLightning/pytorch-lightning/pull/8258))
+
+
 ## [1.3.8] - 2021-07-01
 
 ### Fixed

pytorch_lightning/__about__.py

@@ -1,7 +1,7 @@
 import time
 
 _this_year = time.strftime("%Y")
-__version__ = '1.4.0dev'
+__version__ = '1.4.0rc0'
 __author__ = 'William Falcon et al.'
 __author_email__ = '[email protected]'
 __license__ = 'Apache-2.0'

pytorch_lightning/loops/epoch/evaluation_epoch_loop.py

@@ -37,9 +37,8 @@ def __init__(self) -> None:
         super().__init__()
         self.predictions: Optional[PredictionCollection] = None
         self.dataloader: Optional[Iterator] = None
-        self.dl_max_batches: Optional[int] = None
-        self.dataloader_idx: Optional[int] = None
-        self.num_dataloaders: Optional[int] = None
+        self._dl_max_batches: Optional[int] = None
+        self._num_dataloaders: Optional[int] = None
         self.outputs: List[STEP_OUTPUT] = []
         self.progress = EpochProgress()
 
@@ -54,15 +53,14 @@ def connect(
     @property
     def done(self) -> bool:
         """Returns ``True`` if the current iteration count reaches the number of dataloader batches."""
-        return self.iteration_count >= self.dl_max_batches
+        return self.iteration_count >= self._dl_max_batches
 
     def reset(self) -> None:
         """Resets the loop's internal state."""
         self.iteration_count = 0
         self.predictions = PredictionCollection(self.trainer.global_rank, self.trainer.world_size)
-        self.dl_max_batches = None
-        self.dataloader_idx = None
-        self.num_dataloaders = None
+        self._dl_max_batches = None
+        self._num_dataloaders = None
         self.outputs = []
 
     def on_run_start(
@@ -80,11 +78,9 @@ def on_run_start(
             dl_max_batches: maximum number of batches the dataloader can produce
             num_dataloaders: the total number of dataloaders
         """
-        void(dataloader_iter)
-
-        self.dl_max_batches = dl_max_batches
-        self.dataloader_idx = dataloader_idx
-        self.num_dataloaders = num_dataloaders
+        void(dataloader_iter, dataloader_idx)
+        self._dl_max_batches = dl_max_batches
+        self._num_dataloaders = num_dataloaders
 
     def advance(
         self,
@@ -182,8 +178,8 @@ def on_evaluation_batch_start(self, batch: Any, batch_idx: int, dataloader_idx:
         """
         self.trainer.logger_connector.on_batch_start()
 
-        assert self.num_dataloaders is not None
-        self.trainer.logger_connector.on_evaluation_batch_start(batch, batch_idx, dataloader_idx, self.num_dataloaders)
+        assert self._num_dataloaders is not None
+        self.trainer.logger_connector.on_evaluation_batch_start(batch, batch_idx, dataloader_idx, self._num_dataloaders)
 
         if self.trainer.testing:
             self.trainer.call_hook("on_test_batch_start", batch, batch_idx, dataloader_idx)
@@ -243,8 +239,8 @@ def _build_kwargs(self, batch: Any, batch_idx: int, dataloader_idx: int) -> Dict
         # make dataloader_idx arg in validation_step optional
         step_kwargs = OrderedDict([("batch", batch), ("batch_idx", batch_idx)])
 
-        multiple_val_loaders = not self.trainer.testing and self.num_dataloaders > 1
-        multiple_test_loaders = self.trainer.testing and self.num_dataloaders > 1
+        multiple_val_loaders = not self.trainer.testing and self._num_dataloaders > 1
+        multiple_test_loaders = self.trainer.testing and self._num_dataloaders > 1
 
         if multiple_test_loaders or multiple_val_loaders:
             step_kwargs["dataloader_idx"] = dataloader_idx