returnn 1.20250422.113157__py3-none-any.whl → 1.20250423.105638__py3-none-any.whl

returnn/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: returnn
-Version: 1.20250422.113157
+Version: 1.20250423.105638
 Summary: The RWTH extensible training framework for universal recurrent neural networks
 Home-page: https://github.com/rwth-i6/returnn/
 Author: Albert Zeyer

returnn/_setup_info_generated.py

@@ -1,2 +1,2 @@
-version = '1.20250422.113157'
-long_version = '1.20250422.113157+git.731b3f5'
+version = '1.20250423.105638'
+long_version = '1.20250423.105638+git.767de47'

returnn/torch/engine.py

@@ -82,15 +82,17 @@ class Engine(EngineBase):
         )  # type: Union[int,float,Dict[str,int],NumbersDict]
         self._orig_model = None  # type: Optional[Union[rf.Module, torch.nn.Module]]
         self._pt_model = None  # type: Optional[torch.nn.Module]
-        self._train_step_func = None  # type: Optional[Callable]
-        self._forward_step_func = self.config.typed_value("forward_step")  # type: Optional[Callable]
-        self._forward_step_expected_outputs = None  # type: Optional[TensorDict]
+        self._epoch_start_func: Optional[Callable] = self.config.typed_value("epoch_start")
+        self._epoch_end_func: Optional[Callable] = self.config.typed_value("epoch_end")
+        self._train_step_func: Optional[Callable] = None
+        self._forward_step_func: Optional[Callable] = self.config.typed_value("forward_step")
+        self._forward_step_expected_outputs: Optional[TensorDict] = None
         if self.config.typed_value("model_outputs") is not None:
             self._forward_step_expected_outputs = TensorDict()
             self._forward_step_expected_outputs.update(self.config.typed_value("model_outputs"), auto_convert=True)
         self._save_model_epoch_interval = 1
         self._ignore_param_set: Set[str] = set()  # for the updater and for saving the model checkpoint
-        self._updater = None  # type: Optional[Updater]
+        self._updater: Optional[Updater] = None
 
         self._use_autocast = False
         self._autocast_dtype = None  # type: Optional[str]
@@ -319,6 +321,26 @@ class Engine(EngineBase):
             ]
             print(f"Memory usage ({self._device}):", " ".join(stats), file=log.v1)
 
+    def _on_epoch_start(self, *, dataset_name: str):
+        if self._epoch_start_func:
+            self._epoch_start_func(
+                epoch=self.epoch,
+                step=self.global_train_step,
+                model=self._orig_model,
+                dataset_name=dataset_name,
+                **util.get_fwd_compat_kwargs(),
+            )
+
+    def _on_epoch_end(self, *, dataset_name: str):
+        if self._epoch_end_func:
+            self._epoch_end_func(
+                epoch=self.epoch,
+                step=self.global_train_step,
+                model=self._orig_model,
+                dataset_name=dataset_name,
+                **util.get_fwd_compat_kwargs(),
+            )
+
     def train_epoch(self):
         """
         train one (sub)epoch
@@ -346,6 +368,8 @@ class Engine(EngineBase):
         self._maybe_reset_dev_memory_caches()
         self._reset_dev_memory_stats()
 
+        self._on_epoch_start(dataset_name="train")
+
         if self.config.bool("debug_shell_before_train_loop", False):
             print("debug_shell_before_train_loop", file=log.v1)
             debug_shell(user_ns=locals(), user_global_ns=globals(), exit_afterwards=False)
@@ -564,6 +588,8 @@ class Engine(EngineBase):
 
         self._maybe_report_dev_memory_stats()
 
+        self._on_epoch_end(dataset_name="train")
+
         if self.epoch % self._save_model_epoch_interval == 0 or self.epoch == self._final_epoch:
             if self.model_filename:
                 self._save_model()
@@ -612,6 +638,8 @@ class Engine(EngineBase):
 
             print(f"Evaluating dataset {dataset_name!r}", file=log.v3)
 
+            self._on_epoch_start(dataset_name=dataset_name)
+
             accumulated_losses_dict = NumbersDict()
             accumulated_inv_norm_factors_dict = NumbersDict()
             step_idx = 0
@@ -685,6 +713,8 @@ class Engine(EngineBase):
                 _has_data = torch.tensor([False], device="cpu", dtype=torch.int8)
                 torch.distributed.broadcast(_has_data, src=0)
 
+            self._on_epoch_end(dataset_name=dataset_name)
+
         if not self._torch_distributed_ctx or self._torch_distributed_ctx.rank() == 0:
             print(
                 f"Epoch {self.epoch} evaluation:",

{returnn-1.20250422.113157.dist-info → returnn-1.20250423.105638.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: returnn
-Version: 1.20250422.113157
+Version: 1.20250423.105638
 Summary: The RWTH extensible training framework for universal recurrent neural networks
 Home-page: https://github.com/rwth-i6/returnn/
 Author: Albert Zeyer

{returnn-1.20250422.113157.dist-info → returnn-1.20250423.105638.dist-info}/RECORD

@@ -1,9 +1,9 @@
-returnn/PKG-INFO,sha256=6zi37tT7m8yIZkqfivfYcN-4Pfq9RPDvVa-XXxDychk,5215
+returnn/PKG-INFO,sha256=7FZF3sif_hKT_9vHgl4snASut9Lc4uPAOGJfqaOxr_A,5215
 returnn/__init__.py,sha256=biBtRsM0WZ406vShaeH-9WFoqJ8XwTbn6g0EeFJ7l8E,1012
 returnn/__main__.py,sha256=qBFbuB1yN3adgVM5pXt2-Yq9vorjRNchNPL8kDKx44M,31752
 returnn/__old_mod_loader__.py,sha256=nvsNY-xELdS_IPNkv66Q9Rmvg4dbGW0-EBRDcCmctos,7654
 returnn/__setup__.py,sha256=22kQn2fh11iPM0hLb2Fy5sLmoU1JGvmDxXRYuRgQkwU,4659
-returnn/_setup_info_generated.py,sha256=A7xaVfPcI0O9Cc0mYkkSvcQmYEgzf52QpAfjMkx9SjI,77
+returnn/_setup_info_generated.py,sha256=_2mdcTTEJ__VlmLagOoes0sVhrZN1SeK64wJQZ4crrY,77
 returnn/config.py,sha256=3tmKhB6FnQZaNdtcYsiB61JnEY--iZ2qmJ4yq0b6tE0,29140
 returnn/forward_iface.py,sha256=A_OJiaXsX4MlXQRzST86ylyxSUZbC402PQL1REcqHjM,911
 returnn/learning_rate_control.py,sha256=ZvWryAn_tv9DhV8sh1LV3eE34Yltl3On3mYZAG4hR9s,34684
@@ -207,7 +207,7 @@ returnn/tf/util/open_fst.py,sha256=sZRDw4TbxvhGqpGdUJWy1ebvlZm4_RPhygpRw9uLAOQ,1
 returnn/torch/README.md,sha256=jzJ2FpOHW02vxN69yKaV97C9LI-hmvjBglKfdZXIDdc,85
 returnn/torch/__init__.py,sha256=MHEUyNHB20Vy89uKAqZoj6FxJKF1Gq3HW-i6ra1pNcI,24
 returnn/torch/distributed.py,sha256=skFyutdVztxgTEk3HHJ8S83qRWbNpkNT8Tj16Ic0_hE,6981
-returnn/torch/engine.py,sha256=jJMVR8ltJMfF70-m0HxR4xNS94wNfO4CSAF0jFlB7_g,76771
+returnn/torch/engine.py,sha256=yfqP9jzOH1OjiETqoBh20YOeEaX_kyr_kwPUkhSFxiI,77833
 returnn/torch/updater.py,sha256=GqtBvZpElPVMm0lq84JPl4NVLFFETZAzAbR0rTomSao,28249
 returnn/torch/data/__init__.py,sha256=6cLNEi8KoGI12PF6akN7mI_mtjlx-0hcQAfMYoExwik,132
 returnn/torch/data/extern_data.py,sha256=OSoy3x1KiyiJCr7DfF5uPFAu09We2N2WbA0yo-pYXxM,7601
@@ -253,8 +253,8 @@ returnn/util/sig_proc.py,sha256=Tjz0VOAVyqu2qDCF5HZ1JjALjcFsHcNkcd96WgZeKfE,7265
 returnn/util/task_system.py,sha256=y4sMVXQ25Qd2z0rx03uOlXlkE-jbCYC1Sjfn-XlraVU,26003
 returnn/util/train_proc_manager.py,sha256=Pjht28k6uz6BNQ47uW6Gf880iyq5q4wx7P_K2tmoAM8,3266
 returnn/util/watch_memory.py,sha256=BR5P2kvBN6UI81cE0_1WAA6Hd1SByLbBaiDxvLhPOew,4213
-returnn-1.20250422.113157.dist-info/LICENSE,sha256=ywBD_U2aD4vpuoIgNAsjIGBYydl0tVKll3De0Z8s77c,11041
-returnn-1.20250422.113157.dist-info/METADATA,sha256=6zi37tT7m8yIZkqfivfYcN-4Pfq9RPDvVa-XXxDychk,5215
-returnn-1.20250422.113157.dist-info/WHEEL,sha256=iAkIy5fosb7FzIOwONchHf19Qu7_1wCWyFNR5gu9nU0,91
-returnn-1.20250422.113157.dist-info/top_level.txt,sha256=Lsn4WZc5Pbfk0-xDQOgnFCxOoqxL4CyeM3N1TFbJncw,8
-returnn-1.20250422.113157.dist-info/RECORD,,
+returnn-1.20250423.105638.dist-info/LICENSE,sha256=ywBD_U2aD4vpuoIgNAsjIGBYydl0tVKll3De0Z8s77c,11041
+returnn-1.20250423.105638.dist-info/METADATA,sha256=7FZF3sif_hKT_9vHgl4snASut9Lc4uPAOGJfqaOxr_A,5215
+returnn-1.20250423.105638.dist-info/WHEEL,sha256=iAkIy5fosb7FzIOwONchHf19Qu7_1wCWyFNR5gu9nU0,91
+returnn-1.20250423.105638.dist-info/top_level.txt,sha256=Lsn4WZc5Pbfk0-xDQOgnFCxOoqxL4CyeM3N1TFbJncw,8
+returnn-1.20250423.105638.dist-info/RECORD,,