returnn 1.20250305.155930__py3-none-any.whl → 1.20250317.160550__py3-none-any.whl

returnn/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: returnn
-Version: 1.20250305.155930
+Version: 1.20250317.160550
 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 = '1.20250305.155930'
-long_version = '1.20250305.155930+git.c6cce38'
+version = '1.20250317.160550'
+long_version = '1.20250317.160550+git.b173148'

returnn/frontend/encoder/conformer.py

@@ -294,7 +294,7 @@ class ConformerEncoder(ISeqDownsamplingEncoder):
     def __init__(
         self,
         in_dim: Dim,
-        out_dim: Dim = Dim(512, name="conformer-enc-default-out-dim"),
+        out_dim: Union[Dim, int] = Dim(512, name="conformer-enc-default-out-dim"),
         *,
         num_layers: int,
         input_layer: Optional[Union[ConformerConvSubsample, ISeqDownsamplingEncoder, rf.Module, Any]],
@@ -333,6 +333,11 @@ class ConformerEncoder(ISeqDownsamplingEncoder):
         """
         super().__init__()
 
+        assert isinstance(in_dim, Dim)
+        if isinstance(out_dim, int):
+            out_dim = Dim(out_dim, name="conformer-enc-out-dim")
+        assert isinstance(out_dim, Dim)
+
         self.in_dim = in_dim
         self.out_dim = out_dim
         self.dropout = dropout

returnn/frontend/nested.py

@@ -21,6 +21,7 @@ def mask_nested(
     s: T,
     *,
     mask: Tensor,
+    mask_cpu: Optional[Tensor] = None,
     mask_value: Union[T, Tensor, float, None],
     dim_map: Optional[Dict[Dim, Dim]] = None,
     allow_dim_extension: bool = True,
@@ -30,6 +31,7 @@ def mask_nested(
 
     :param s:
     :param mask:
+    :param mask_cpu: mask tensor for CPU. this is used e.g. for dyn dim sizes
     :param mask_value:
     :param dim_map:
     :param allow_dim_extension:
@@ -37,7 +39,7 @@ def mask_nested(
     """
     if dim_map is None:
         dim_map = {}
-    partial_kwargs = dict(mask=mask, dim_map=dim_map, allow_dim_extension=allow_dim_extension)
+    partial_kwargs = dict(mask=mask, mask_cpu=mask_cpu, dim_map=dim_map, allow_dim_extension=allow_dim_extension)
     structures = [s]
     if type(s) is type(mask_value):  # mask_value also same nested structure?
         tree.assert_same_structure(s, mask_value)
@@ -49,7 +51,13 @@ def mask_nested(
 
 
 def _mask_prepare_dims(
-    s: T, mask_value: Union[T, Tensor, float, None], *, mask: Tensor, dim_map: Dict[Dim, Dim], allow_dim_extension: bool
+    s: T,
+    mask_value: Union[T, Tensor, float, None],
+    *,
+    mask: Tensor,
+    mask_cpu: Optional[Tensor] = None,
+    dim_map: Dict[Dim, Dim],
+    allow_dim_extension: bool,
 ) -> T:
     if isinstance(s, Dim):
         if mask_value is None:
@@ -69,6 +77,7 @@ def _mask_prepare_dims(
         new_dyn_size = _mask(
             s.get_size_tensor(),
             mask=mask,
+            mask_cpu=mask_cpu,
             mask_value=mask_value.get_size_tensor(),
             dim_map=dim_map,
             allow_dim_extension=allow_dim_extension,
@@ -80,11 +89,19 @@ def _mask_prepare_dims(
 
 
 def _mask(
-    s: T, mask_value: Union[T, Tensor, float, None], *, mask: Tensor, dim_map: Dict[Dim, Dim], allow_dim_extension: bool
+    s: T,
+    mask_value: Union[T, Tensor, float, None],
+    *,
+    mask: Tensor,
+    mask_cpu: Optional[Tensor] = None,
+    dim_map: Dict[Dim, Dim],
+    allow_dim_extension: bool,
 ) -> T:
     if s is None:
         return s
     if isinstance(s, Tensor):
+        if s.device == "cpu" and mask_cpu is not None:
+            mask = mask_cpu
         if dim_map:
             for d in s.dims:
                 if d in dim_map:

{returnn-1.20250305.155930.dist-info → returnn-1.20250317.160550.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: returnn
-Version: 1.20250305.155930
+Version: 1.20250317.160550
 Summary: The RWTH extensible training framework for universal recurrent neural networks
 Home-page: https://github.com/rwth-i6/returnn/
 Author: Albert Zeyer

{returnn-1.20250305.155930.dist-info → returnn-1.20250317.160550.dist-info}/RECORD

@@ -1,9 +1,9 @@
-returnn/PKG-INFO,sha256=QrtdJEd1UUGZh5SpkBvrgoXEka1vXc8i6Gcjd7EkQ9U,5215
+returnn/PKG-INFO,sha256=TV21aoMQ025cRAZICtmi0RJ_meXE0a64AGfhceYHZjI,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=-Ag7t9l0Rop5Xdrb4ipMF5x6UbTBss6lGxGITzTntCA,77
+returnn/_setup_info_generated.py,sha256=ArtH4g8hmYryE0Dl4mKk7JFa1Qkulave6EJ1xHrafOY,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
@@ -104,7 +104,7 @@ returnn/frontend/loss.py,sha256=r_qRAiFIXgXTnFuLoAhE5jsyAZYMCucRzf55XfWbkC8,7441
 returnn/frontend/math_.py,sha256=KlJxdIib8ENlid7cc4lcwHv5e21tzTjTEV8VgEDAijo,16984
 returnn/frontend/matmul.py,sha256=3QaGiZtSs9PriT40T7Vc3KnYKPgYSN4tCZytYeq9qMA,1945
 returnn/frontend/module.py,sha256=219rh5mE0CD0-NdxXLsKyhv3BNtOI9jSyiI1Rb8MOyU,10700
-returnn/frontend/nested.py,sha256=oDiqnyTML7ZtCxrufU4ypG0fOZ_WsZPIvfdSn1Phs6M,14698
+returnn/frontend/nested.py,sha256=Hm4GT5ZI1OyWpYxv_SP5jlBztJsjGVMgtvKJnvQYa00,15068
 returnn/frontend/normalization.py,sha256=QIjXYg0C8BD2g_1lAkVO4Cara729uHC_bsQh99VsWeI,14061
 returnn/frontend/parameter.py,sha256=w6SN-uv87OyeWBt90_3UBbK0h6sftSOCxkqXPg76caY,10375
 returnn/frontend/parametrizations.py,sha256=hVbOlgm1pQAmZnAnNxq8Tk23rykr_iy3-6R1H6CwlMA,2798
@@ -138,7 +138,7 @@ returnn/frontend/decoder/__init__.py,sha256=A-koKyPVlXp_V_2bk6GKZ1Xfv4rYIcfxGMXQ
 returnn/frontend/decoder/transformer.py,sha256=GzZ-1HxDjmzM8bsW9KZn-4_W97vsnChbo_NtL-OBYJA,23179
 returnn/frontend/encoder/__init__.py,sha256=0QGLlujRIKx3zBREeShza_-xhGIxj73zbd7t-g1m-ho,17
 returnn/frontend/encoder/base.py,sha256=A759EwCYAmSi-kzXz1vaTjR2l59TvNGQlzaNdp3UOKs,2109
-returnn/frontend/encoder/conformer.py,sha256=YPtH0Clx2QrKOoxbtUSkYR7QiDp-EYmoOcGc_gc2ZEk,19778
+returnn/frontend/encoder/conformer.py,sha256=ro0uzEzDbAyNGYN5ff0KmiDl4HOYQluu64mJxYzuy-M,19972
 returnn/frontend/encoder/conformer_v2.py,sha256=vAYdT8m2Zzg3IIZZafeccClFHU1_c9T-EgBOsHadQPA,7701
 returnn/frontend/encoder/e_branchformer.py,sha256=zEla-iXJciK7bCenlTwsPB8dXo_VPMlFm2xc3op_lPY,12278
 returnn/frontend/encoder/transformer.py,sha256=Jj0mF1D2MohOk-9sGYdsLtVW_86fwoq4pKWCdPMvPR8,11580
@@ -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-1.20250305.155930.dist-info/LICENSE,sha256=ywBD_U2aD4vpuoIgNAsjIGBYydl0tVKll3De0Z8s77c,11041
-returnn-1.20250305.155930.dist-info/METADATA,sha256=QrtdJEd1UUGZh5SpkBvrgoXEka1vXc8i6Gcjd7EkQ9U,5215
-returnn-1.20250305.155930.dist-info/WHEEL,sha256=P9jw-gEje8ByB7_hXoICnHtVCrEwMQh-630tKvQWehc,91
-returnn-1.20250305.155930.dist-info/top_level.txt,sha256=Lsn4WZc5Pbfk0-xDQOgnFCxOoqxL4CyeM3N1TFbJncw,8
-returnn-1.20250305.155930.dist-info/RECORD,,
+returnn-1.20250317.160550.dist-info/LICENSE,sha256=ywBD_U2aD4vpuoIgNAsjIGBYydl0tVKll3De0Z8s77c,11041
+returnn-1.20250317.160550.dist-info/METADATA,sha256=TV21aoMQ025cRAZICtmi0RJ_meXE0a64AGfhceYHZjI,5215
+returnn-1.20250317.160550.dist-info/WHEEL,sha256=iAkIy5fosb7FzIOwONchHf19Qu7_1wCWyFNR5gu9nU0,91
+returnn-1.20250317.160550.dist-info/top_level.txt,sha256=Lsn4WZc5Pbfk0-xDQOgnFCxOoqxL4CyeM3N1TFbJncw,8
+returnn-1.20250317.160550.dist-info/RECORD,,

{returnn-1.20250305.155930.dist-info → returnn-1.20250317.160550.dist-info}/WHEEL

@@ -1,5 +1,5 @@
 Wheel-Version: 1.0
-Generator: setuptools (75.3.0)
+Generator: setuptools (75.3.2)
 Root-Is-Purelib: true
 Tag: py3-none-any