Log batch metrics

CHANGELOG.md

@@ -25,6 +25,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   model's `__init__()` method by wrapping them with `self.ddp_accelerator.wrap_module()`. See the `allennlp.modules.transformer.t5`
   for an example.
 - Added Tango components, to be explored in detail in a later post.
+- We now log batch metrics to tensorboard and wandb.
 
 ### Fixed
 
@@ -34,15 +35,16 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - `TransformerTextField` can now take tensors of shape `(1, n)` like the tensors produced from a HuggingFace tokenizer.
 - `tqdm` lock is now set inside `MultiProcessDataLoading` when new workers are spawned to avoid contention when writing output.
 - `ConfigurationError` is now pickleable.
-- Multitask models now support `TextFieldTensor` in heads, not just in the backbone
+- Multitask models now support `TextFieldTensor` in heads, not just in the backbone.
 
 ### Changed
 
 - The type of the `grad_norm` parameter of `GradientDescentTrainer` is now `Union[float, bool]`,
   with a default value of `False`. `False` means gradients are not rescaled and the gradient
   norm is never even calculated. `True` means the gradients are still not rescaled but the gradient
   norm is calculated and passed on to callbacks. A `float` value means gradients are rescaled.
-- `TensorCache` now supports more concurrent readers and writers. 
+- `TensorCache` now supports more concurrent readers and writers.
+- We no longer log parameter statistics to tensorboard or wandb by default.
 
 
 ## [v2.6.0](https://github.com/allenai/allennlp/releases/tag/v2.6.0) - 2021-07-19

allennlp/training/callbacks/log_writer.py

@@ -227,7 +227,7 @@ def log_batch(
 
             # Now collect per-batch metrics to log.
             metrics_to_log: Dict[str, float] = {}
-            for key in ("batch_loss", "batch_reg_loss"):
+            for key in ["batch_loss", "batch_reg_loss"]:
                 if key not in metrics:
                     continue
                 value = metrics[key]
@@ -241,6 +241,13 @@ def log_batch(
                     self._batch_loss_moving_items[key]
                 )
 
+            for key, value in metrics.items():
+                if key.startswith("batch_"):
+                    continue
+                key = "batch_" + key
+                if key not in metrics_to_log:
+                    metrics_to_log[key] = value
+
             self.log_scalars(
                 metrics_to_log,
                 log_prefix="train",
@@ -253,7 +260,7 @@ def log_batch(
 
         if self._batch_size_interval:
             # We're assuming here that `log_batch` will get called every batch, and only every
-            # batch.  This is true with our current usage of this code (version 1.0); if that
+            # batch. This is true with our current usage of this code (version 1.0); if that
             # assumption becomes wrong, this code will break.
             batch_group_size = sum(get_batch_size(batch) for batch in batch_group)  # type: ignore
             self._cumulative_batch_group_size += batch_group_size

allennlp/training/callbacks/tensorboard.py

@@ -21,7 +21,7 @@ def __init__(
         summary_interval: int = 100,
         distribution_interval: Optional[int] = None,
         batch_size_interval: Optional[int] = None,
-        should_log_parameter_statistics: bool = True,
+        should_log_parameter_statistics: bool = False,
         should_log_learning_rate: bool = False,
     ) -> None:
         super().__init__(