Merge branch 'main' into tp-tests

composer/utils/remote_uploader.py

@@ -92,7 +92,10 @@ def upload_file(retry_index: int = 0):
                 # Good! It shouldn't exist.
                 pass
             else:
-                raise FileExistsError(f'Object {remote_file_name} already exists, but overwrite was set to False.')
+                raise FileExistsError(
+                    f'Object {remote_file_name} already exists, but overwrite was set to False. '
+                    'Please set `save_overwrite` to `True` in Trainer to overwrite the existing file.',
+                )
         log.info(f'Uploading file {local_file_path} to {remote_file_name}')
         object_store.upload_object(
             object_name=remote_file_name,

setup.py

@@ -160,7 +160,7 @@ def package_files(prefix: str, directory: str, extension: str):
 ]
 
 extra_deps['wandb'] = [
-    'wandb>=0.13.2,<0.18',
+    'wandb>=0.13.2,<0.19',
 ]
 
 extra_deps['comet_ml'] = [

tests/loggers/test_wandb_logger.py

@@ -269,10 +269,10 @@ def test_wandb_log_metrics(test_wandb_logger):
     eval_metrics_cross_entropy_count = all_run_text.count('metrics/eval/CrossEntropy')
     train_loss_count = all_run_text.count('loss/train/total')
 
-    expected_number_train_loss_count = (dataset_size / batch_size) + 1  # wandb includes it in the file one extra time
+    expected_number_train_loss_count = (dataset_size / batch_size) * 2  # wandb includes it twice per step
     expected_number_train_metrics_count = (
         dataset_size / batch_size
-    ) + 2  # wandb includes it in the file two extra times
+    ) * 2 + 2  # wandb includes it twice per step plus two extra times
     expected_number_eval_metrics_count = 2  # wandb includes it in the file twice
     assert train_metrics_accuracy_count == expected_number_train_metrics_count
     assert train_loss_count == expected_number_train_loss_count