returnn 1.20250211.215623__py3-none-any.whl → 1.20250216.155246__py3-none-any.whl

returnn/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: returnn
-Version: 1.20250211.215623
+Version: 1.20250216.155246
 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 = '1.20250211.215623'
-long_version = '1.20250211.215623+git.8640885'
+version = '1.20250216.155246'
+long_version = '1.20250216.155246+git.1355f14'

returnn/frontend/audio/specaugment.py

@@ -44,25 +44,27 @@ def specaugment(
         x_masked = x
         spatial_len = spatial_dim.get_dim_value_tensor()
         # time mask
-        x_masked = random_mask(
-            x_masked,
-            mask_axis=spatial_dim,
-            broadcast_axis=feature_dim,
-            min_num=rf.minimum(step1 + step2, spatial_len),
-            max_num=rf.minimum(
-                rf.maximum(spatial_len // num_spatial_mask_factor, 2) * (step0 + step1 + step2 * 2), spatial_len
-            ),
-            max_dims=max_consecutive_spatial_dims,
-        )
+        if max_consecutive_spatial_dims > 0 and num_spatial_mask_factor > 0:
+            x_masked = random_mask(
+                x_masked,
+                mask_axis=spatial_dim,
+                broadcast_axis=feature_dim,
+                min_num=rf.minimum(step1 + step2, spatial_len),
+                max_num=rf.minimum(
+                    rf.maximum(spatial_len // num_spatial_mask_factor, 2) * (step0 + step1 + step2 * 2), spatial_len
+                ),
+                max_dims=max_consecutive_spatial_dims,
+            )
         # feature mask
-        x_masked = random_mask(
-            x_masked,
-            mask_axis=feature_dim,
-            broadcast_axis=spatial_dim,
-            min_num=step1 + step2,
-            max_num=step0 * 2 + step1 + step2 * 2,
-            max_dims=max_consecutive_feature_dims,
-        )
+        if max_consecutive_feature_dims > 0:
+            x_masked = random_mask(
+                x_masked,
+                mask_axis=feature_dim,
+                broadcast_axis=spatial_dim,
+                min_num=step1 + step2,
+                max_num=step0 * 2 + step1 + step2 * 2,
+                max_dims=max_consecutive_feature_dims,
+            )
         return x_masked
 
     return rf.cond(rf.get_run_ctx().train_flag | (not only_on_train), _mask_branch, lambda: x)

returnn/torch/frontend/_backend.py

@@ -1477,6 +1477,10 @@ class TorchBackend(Backend[torch.Tensor]):
             elif mode == "mean":
                 mask_value = 0
                 correction_factor = rf.masked_fraction_of_shape(axis, inverse=True)
+            elif mode == "all":
+                mask_value = True
+            elif mode == "any":
+                mask_value = False
             else:
                 raise NotImplementedError(f"reduce_{mode} not implemented with masking on tensor {source!r}.")
             for dim in axis:

{returnn-1.20250211.215623.dist-info → returnn-1.20250216.155246.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: returnn
-Version: 1.20250211.215623
+Version: 1.20250216.155246
 Summary: The RWTH extensible training framework for universal recurrent neural networks
 Home-page: https://github.com/rwth-i6/returnn/
 Author: Albert Zeyer

{returnn-1.20250211.215623.dist-info → returnn-1.20250216.155246.dist-info}/RECORD

@@ -1,9 +1,9 @@
-returnn/PKG-INFO,sha256=1no-Aukog5H0Mvgs1485nTteX29Fjp-sH7_kGvbMHG4,5215
+returnn/PKG-INFO,sha256=9t6Xs3We7RWGbRVGOnZBXT4SZd2GGRmGv-WEAUk64SU,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=xvdH59j45tyParGsCGfKR4RXzo9TQAImKpm20b4ylGk,77
+returnn/_setup_info_generated.py,sha256=8r3qFPJ-iC6jC5sAl9bHpms0G1qv_QL7xYo9i8uhyH4,77
 returnn/config.py,sha256=3tmKhB6FnQZaNdtcYsiB61JnEY--iZ2qmJ4yq0b6tE0,29140
 returnn/forward_iface.py,sha256=A_OJiaXsX4MlXQRzST86ylyxSUZbC402PQL1REcqHjM,911
 returnn/learning_rate_control.py,sha256=ZvWryAn_tv9DhV8sh1LV3eE34Yltl3On3mYZAG4hR9s,34684
@@ -129,7 +129,7 @@ returnn/frontend/_native/tensor_ops.cpp,sha256=G_geJq_9yaTYoG7YsZydiTTkJoqdYB-D8
 returnn/frontend/_native/tensor_ops.hpp,sha256=dDqvUejRNHjItnmOP5aHyAQbAmXmXoDVXSe3tveEU8A,3732
 returnn/frontend/audio/__init__.py,sha256=8mahwucBje8qHKw0bOvoySlvvD0rFKxviSvcAHSjiJY,67
 returnn/frontend/audio/mel.py,sha256=VZdxf2mTLzLOXsLRzCvaad712Zf0c2iwdthrzeVfgxk,7885
-returnn/frontend/audio/specaugment.py,sha256=6GvVkLdRvwyrRkBr0E0kzvbX22W5e6INDQDMWDkOJJY,5604
+returnn/frontend/audio/specaugment.py,sha256=nw8PepKPPwmI13-QyGBm45QjnoY4I_FEjA-_X6KIwzM,5798
 returnn/frontend/conversions/__init__.py,sha256=7plsDxWVYhASa-3qmqbdzSI34A9ujUH2iMkL3eRD0TI,84
 returnn/frontend/conversions/espnet_e_branchformer.py,sha256=Mmp3G6nySy0CqeHa-um-RAuUSnFH1DKNjBbqQB_Pomo,9018
 returnn/frontend/conversions/hf_llama.py,sha256=1WQOhQyUWwkAznaRqK2zpThP8XZbaomkaE8qMG_bZPY,9662
@@ -216,7 +216,7 @@ returnn/torch/data/queued_data_iter.py,sha256=PoOsGHdHVZjTmcyfq_ZOw--P6hyfTdmAWI
 returnn/torch/data/returnn_dataset_wrapper.py,sha256=1Bw82-Ge_8m_DSDXZNqQ3zGDic2HQlp6jysELL0NVK0,7369
 returnn/torch/data/tensor_utils.py,sha256=-Teqi--LLbt6q_5mDRdoHZHmPgSdC83W706ukif_YiU,1284
 returnn/torch/frontend/__init__.py,sha256=AA48HZnC17ASuKA0EWy8loZ-Bib_yUtqF4T1wYvjst4,62
-returnn/torch/frontend/_backend.py,sha256=h_rUhBPxLRgpZSqX4C8vX8q4dHWMhZpwPmGbKN6MsZo,99995
+returnn/torch/frontend/_backend.py,sha256=sdC4kj7_2wyEVxX536TunbY4VbFaScSnMIBi1F27bsE,100128
 returnn/torch/frontend/_rand.py,sha256=1JgIkV2XmpgJD86zXZ-NCAe-QuoP2swr6NaS1oz3Qa8,1830
 returnn/torch/frontend/bridge.py,sha256=Z2_UW8AagezC7zsXDc5PKcd8G9WwisV7j9SWGHU0m4U,7840
 returnn/torch/frontend/raw_ops.py,sha256=lF0h-KtYYsdaaqQADylVZp9qzPskOOXA4MfmYDyx5IU,296
@@ -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-1.20250211.215623.dist-info/LICENSE,sha256=ywBD_U2aD4vpuoIgNAsjIGBYydl0tVKll3De0Z8s77c,11041
-returnn-1.20250211.215623.dist-info/METADATA,sha256=1no-Aukog5H0Mvgs1485nTteX29Fjp-sH7_kGvbMHG4,5215
-returnn-1.20250211.215623.dist-info/WHEEL,sha256=P9jw-gEje8ByB7_hXoICnHtVCrEwMQh-630tKvQWehc,91
-returnn-1.20250211.215623.dist-info/top_level.txt,sha256=Lsn4WZc5Pbfk0-xDQOgnFCxOoqxL4CyeM3N1TFbJncw,8
-returnn-1.20250211.215623.dist-info/RECORD,,
+returnn-1.20250216.155246.dist-info/LICENSE,sha256=ywBD_U2aD4vpuoIgNAsjIGBYydl0tVKll3De0Z8s77c,11041
+returnn-1.20250216.155246.dist-info/METADATA,sha256=9t6Xs3We7RWGbRVGOnZBXT4SZd2GGRmGv-WEAUk64SU,5215
+returnn-1.20250216.155246.dist-info/WHEEL,sha256=P9jw-gEje8ByB7_hXoICnHtVCrEwMQh-630tKvQWehc,91
+returnn-1.20250216.155246.dist-info/top_level.txt,sha256=Lsn4WZc5Pbfk0-xDQOgnFCxOoqxL4CyeM3N1TFbJncw,8
+returnn-1.20250216.155246.dist-info/RECORD,,