returnn 1.20251013.113026__py3-none-any.whl → 1.20251013.131953__py3-none-any.whl

returnn/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: returnn
-Version: 1.20251013.113026
+Version: 1.20251013.131953
 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 = '1.20251013.113026'
-long_version = '1.20251013.113026+git.dbc7e8e'
+version = '1.20251013.131953'
+long_version = '1.20251013.131953+git.e1762d8'

returnn/torch/engine.py

@@ -134,6 +134,14 @@ class Engine(EngineBase):
         self._forward_auto_split_batch_on_oom = config.bool("forward_auto_split_batch_on_oom", False)
         self._stop_on_nonfinite_train_score = config.bool("stop_on_nonfinite_train_score", True)
 
+        if config.bool("use_tensorboard", False):
+            from torch.utils.tensorboard import SummaryWriter
+
+            self._tensorboard_writer = SummaryWriter()
+            self._tensorboard_opts = config.typed_value("tensorboard_opts", {})
+        else:
+            self._tensorboard_writer = None
+
         default_float_dtype = config.value("default_float_dtype", None)
         if default_float_dtype is not None:
             assert isinstance(default_float_dtype, str)
@@ -257,6 +265,9 @@ class Engine(EngineBase):
             self.init_train_epoch()
             self.train_epoch()
 
+        if self._tensorboard_writer:
+            self._tensorboard_writer.close()
+
         print(f"Finished training at epoch {self.epoch}, global train step {self.global_train_step}", file=log.v3)
 
     def init_train_epoch(self):
@@ -513,6 +524,18 @@ class Engine(EngineBase):
                     batch_size_info=_get_batch_size_info(extern_data) if self._log_batch_size else None,
                     log_memory_usage_device=self._device if self._log_memory_usage else None,
                 )
+                if (
+                    self._tensorboard_writer
+                    and self.global_train_step % self._tensorboard_opts.get("log_every_n_train_steps", 100) == 0
+                ):
+                    # write losses/errors to tensorboard
+                    for key, val in eval_info.items():
+                        self._tensorboard_writer.add_scalar(f"train/{key}", val, global_step=self.global_train_step)
+                    self._tensorboard_writer.add_scalar(
+                        f"train/learning_rate",
+                        self._updater.get_effective_learning_rate(),
+                        global_step=self.global_train_step,
+                    )
 
                 if self._stop_on_nonfinite_train_score:
                     if any(np.isinf(v) or np.isnan(v) for v in accumulated_losses_dict.values()):
@@ -702,12 +725,20 @@ class Engine(EngineBase):
                         start_elapsed=step_end_time - eval_start_time,
                         log_memory_usage_device=self._device if self._log_memory_usage else None,
                     )
+
                     step_idx += 1
 
             assert step_idx > 0, f"No data in dataset {dataset_name!r}."
             accumulated_losses_dict = accumulated_losses_dict / accumulated_inv_norm_factors_dict
             accumulated_losses_dict = self._maybe_extend_losses_info(accumulated_losses_dict)
 
+            if self._tensorboard_writer:
+                # write losses/errors to tensorboard
+                for key, val in accumulated_losses_dict.items():
+                    self._tensorboard_writer.add_scalar(
+                        f"{dataset_name}/{key}", val, global_step=self.global_train_step
+                    )
+
             self.learning_rate_control.set_epoch_error(
                 self.epoch, {f"{dataset_name}_loss_{k}": v for k, v in accumulated_losses_dict.items()}
             )

{returnn-1.20251013.113026.dist-info → returnn-1.20251013.131953.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: returnn
-Version: 1.20251013.113026
+Version: 1.20251013.131953
 Summary: The RWTH extensible training framework for universal recurrent neural networks
 Home-page: https://github.com/rwth-i6/returnn/
 Author: Albert Zeyer

{returnn-1.20251013.113026.dist-info → returnn-1.20251013.131953.dist-info}/RECORD

@@ -1,9 +1,9 @@
-returnn/PKG-INFO,sha256=21rH_SscXsPoNZbOjOnE6SMzkvwNNZWKTFzl8ZkxEmI,5215
+returnn/PKG-INFO,sha256=gol7200kSqRcGRCYmHX_Vfg0o3XeXoixOD4si9dZQpM,5215
 returnn/__init__.py,sha256=biBtRsM0WZ406vShaeH-9WFoqJ8XwTbn6g0EeFJ7l8E,1012
 returnn/__main__.py,sha256=lHyZcu_0yc9f7Vf_Kfdy9PmeU0T76XVXnpalHi5WKro,31740
 returnn/__old_mod_loader__.py,sha256=nvsNY-xELdS_IPNkv66Q9Rmvg4dbGW0-EBRDcCmctos,7654
 returnn/__setup__.py,sha256=22kQn2fh11iPM0hLb2Fy5sLmoU1JGvmDxXRYuRgQkwU,4659
-returnn/_setup_info_generated.py,sha256=UBxhUB_ZDJrxS3j3Z2ZbXF6eKFLL9jhpqCApzvEmhr8,77
+returnn/_setup_info_generated.py,sha256=mhKUcDNacm00cBntcCP2-qZd0QlSy6jmbw7ILyfu7dQ,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
@@ -208,7 +208,7 @@ 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=_lyJR71HIoCHpMi5GztGM7YwrX54Am8zSkjnDkE1Lbk,7524
-returnn/torch/engine.py,sha256=JSsQZZiVs9TxRyFEJuR3iH-YZb9sRw7TzoIAIqmplZY,78275
+returnn/torch/engine.py,sha256=Lj_Go3Q5oHOVrHdNcz_6YLjODsiGUsJVMbU0EzSDByk,79677
 returnn/torch/updater.py,sha256=nNd1mBPQyvIB096BEFi0KKmRI-U3jnRETzb743p2B9c,32064
 returnn/torch/data/__init__.py,sha256=6cLNEi8KoGI12PF6akN7mI_mtjlx-0hcQAfMYoExwik,132
 returnn/torch/data/extern_data.py,sha256=5al706ZaYtHWLp5VH2vS-rW69YXP3NHyOFRKY0WY714,7810
@@ -254,8 +254,8 @@ 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-1.20251013.113026.dist-info/LICENSE,sha256=ywBD_U2aD4vpuoIgNAsjIGBYydl0tVKll3De0Z8s77c,11041
-returnn-1.20251013.113026.dist-info/METADATA,sha256=21rH_SscXsPoNZbOjOnE6SMzkvwNNZWKTFzl8ZkxEmI,5215
-returnn-1.20251013.113026.dist-info/WHEEL,sha256=iAkIy5fosb7FzIOwONchHf19Qu7_1wCWyFNR5gu9nU0,91
-returnn-1.20251013.113026.dist-info/top_level.txt,sha256=Lsn4WZc5Pbfk0-xDQOgnFCxOoqxL4CyeM3N1TFbJncw,8
-returnn-1.20251013.113026.dist-info/RECORD,,
+returnn-1.20251013.131953.dist-info/LICENSE,sha256=ywBD_U2aD4vpuoIgNAsjIGBYydl0tVKll3De0Z8s77c,11041
+returnn-1.20251013.131953.dist-info/METADATA,sha256=gol7200kSqRcGRCYmHX_Vfg0o3XeXoixOD4si9dZQpM,5215
+returnn-1.20251013.131953.dist-info/WHEEL,sha256=iAkIy5fosb7FzIOwONchHf19Qu7_1wCWyFNR5gu9nU0,91
+returnn-1.20251013.131953.dist-info/top_level.txt,sha256=Lsn4WZc5Pbfk0-xDQOgnFCxOoqxL4CyeM3N1TFbJncw,8
+returnn-1.20251013.131953.dist-info/RECORD,,