Checkpoint is up to date when resuming training (#1043)

CHANGELOG.md

@@ -11,6 +11,12 @@ Note that Sockeye has checks in place to not translate with an old model that wa
 
 Each version section may have subsections for: _Added_, _Changed_, _Removed_, _Deprecated_, and _Fixed_.
 
+## [3.1.11]
+
+### Fixed
+
+- When resuming training with a fully trained model, `sockeye-train` will correctly exit without creating a duplicate (but separately numbered) checkpoint.
+
 ## [3.1.10]
 
 ### Fixed

sockeye/__init__.py

@@ -11,4 +11,4 @@
 # express or implied. See the License for the specific language governing
 # permissions and limitations under the License.
 
-__version__ = '3.1.10'
+__version__ = '3.1.11'

sockeye/training.py

@@ -209,7 +209,9 @@ def fit(self,
                         self.config.max_updates,
                         self.config.max_checkpoints)
 
-        checkpoint_up_to_date = False
+        # At the start of training, the checkpoint is only up to date if it has
+        # just been loaded (resuming training with an existing model directory).
+        checkpoint_up_to_date = resume_training
         while True:
             if self.config.max_epochs is not None and self.state.epoch == self.config.max_epochs:
                 logger.info("Maximum # of epochs (%s) reached.", self.config.max_epochs)