Replace automatic nan check with optional flag (#1475)

* Replace automatic nan check with optional flag

* Update CHANGELOG.md

CHANGELOG.md

@@ -9,13 +9,13 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 ### Added
 
 - Added `auto_select_gpus` flag to trainer that enables automatic selection of available GPUs on exclusive mode systems.
-- Added learining rate finder ([#1347](https://github.com/PyTorchLightning/pytorch-lightning/pull/1347))
+- Added learning rate finder ([#1347](https://github.com/PyTorchLightning/pytorch-lightning/pull/1347))
 
-- 
+- Added `terminate_on_nan` flag to trainer that performs a NaN check with each training iteration when set to `True`. ([#1475](https://github.com/PyTorchLightning/pytorch-lightning/pull/1475))
 
 ### Changed
 
-- 
+- Changed the default behaviour to no longer include a NaN check with each training iteration. ([#1475](https://github.com/PyTorchLightning/pytorch-lightning/pull/1475))
 
 - 
 

pytorch_lightning/trainer/trainer.py

@@ -134,6 +134,7 @@ class Trainer(
             use_amp=None,  # backward compatible, todo: remove in v0.9.0
             show_progress_bar=None,  # backward compatible, todo: remove in v0.9.0
             nb_sanity_val_steps=None,  # backward compatible, todo: remove in v0.8.0
+            terminate_on_nan: bool = False,
             **kwargs
     ):
         r"""
@@ -281,6 +282,9 @@ class Trainer(
                 To use a different key, set a string instead of True with the key name.
 
             benchmark: If true enables cudnn.benchmark.
+
+            terminate_on_nan: If set to True, will terminate training (by raising a `ValueError`) at the
+                end of each training batch, if any of the parameters or the loss are NaN or +/-inf.
         """
 
         # Init callbacks
@@ -357,6 +361,7 @@ class Trainer(
 
         self.truncated_bptt_steps = truncated_bptt_steps
         self.resume_from_checkpoint = resume_from_checkpoint
+        self.terminate_on_nan = terminate_on_nan
         self.shown_warnings = set()
 
         self.fast_dev_run = fast_dev_run

pytorch_lightning/trainer/training_loop.py

@@ -121,7 +121,8 @@ When this flag is enabled each batch is split into sequences of size truncated_b
 
 NaN detection and intervention
 ------------------------------
-In every forward pass in training, Lightning will check that
+When the `terminate_on_nan` flag is enabled, after every forward pass during training, Lightning will
+check that
 
 1. the loss you return in `training_step` is finite (not NaN and not +/-inf)
 2. the model parameters have finite values.
@@ -130,6 +131,14 @@ Lightning will terminate the training loop with an error message if NaN or infin
 values are detected. If this happens, you should investigate numerically unstable operations
 in your model.
 
+.. code-block:: python
+
+    # DEFAULT (won't perform the NaN check)
+    trainer = Trainer(terminate_on_nan=False)
+
+    # (NaN check each batch and terminate on NaN or infinite values)
+    trainer = Trainer(terminate_on_nan=True)
+
 """
 
 import copy
@@ -216,6 +225,7 @@ class TrainerTrainLoopMixin(ABC):
     min_steps: int
     total_batch_idx: int
     checkpoint_callback: ...
+    terminate_on_nan: bool
 
     # Callback system
     callbacks: List[Callback]
@@ -604,7 +614,8 @@ class TrainerTrainLoopMixin(ABC):
                 loss, batch_output = optimizer_closure()
 
                 # check if loss or model weights are nan
-                self.detect_nan_tensors(loss)
+                if self.terminate_on_nan:
+                    self.detect_nan_tensors(loss)
 
                 # track total loss for logging (avoid mem leaks)
                 self.batch_loss_value.append(loss)

tests/trainer/test_trainer.py

@@ -593,6 +593,7 @@ def test_nan_loss_detection(tmpdir):
     trainer = Trainer(
         default_root_dir=tmpdir,
         max_steps=(test_step + 1),
+        terminate_on_nan=True
     )
 
     with pytest.raises(ValueError, match=r'.*The loss returned in `training_step` is nan or inf.*'):
@@ -619,6 +620,7 @@ def test_nan_params_detection(tmpdir):
     trainer = Trainer(
         default_root_dir=tmpdir,
         max_steps=(test_step + 1),
+        terminate_on_nan=True
     )
 
     with pytest.raises(ValueError, match=r'.*Detected nan and/or inf values in `c_d1.bias`.*'):