returnn 1.20230816.14926__tar.gz → 1.20230816.20915__tar.gz

{returnn-1.20230816.14926/returnn.egg-info → returnn-1.20230816.20915}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: returnn
-Version: 1.20230816.14926
+Version: 1.20230816.20915
 Summary: The RWTH extensible training framework for universal recurrent neural networks
 Home-page: https://github.com/rwth-i6/returnn/
 Author: Albert Zeyer

returnn-1.20230816.20915/_setup_info_generated.py

@@ -0,0 +1,2 @@
+version = '1.20230816.020915'
+long_version = '1.20230816.020915+git.37ebe2e'

{returnn-1.20230816.14926 → returnn-1.20230816.20915}/returnn/frontend/loop.py

@@ -146,7 +146,7 @@ def scan(
             seq_len_ = seq_len_ + rf.cast(prev_cond, dtype=seq_len_.dtype)
             y, s = body(None, s)
             tree.assert_same_structure(ys_, y)
-            ys_ = tree.map_structure(lambda ys__, y_: ys__.push_back(y_), ys_, y)
+            ys_ = tree.map_structure(lambda ys__, y_: ys__.push_back(y_) if ys__ is not None else None, ys_, y)
             c = cond(None, s)
             c = rf.logical_and(c, prev_cond)
             return i + 1, seq_len_, c, s, ys_
@@ -168,7 +168,7 @@ def scan(
                 rf.constant(0, dtype=rf.get_default_array_index_dtype(), dims=cond_dims),  # seq_len
                 initial_cond,  # initial cond. keep this in state such that we can update seq_len in body
                 initial,  # state
-                tree.map_structure(lambda y: TensorArray(y), ys),
+                tree.map_structure(lambda y: TensorArray(y) if y is not None else None, ys),
             ),
         )
 
@@ -188,7 +188,7 @@ def scan(
             i, s, ys_ = _s
             y, s = body(tree.map_structure(lambda x: x[i], xs), s)
             tree.assert_same_structure(ys_, y)
-            ys_ = tree.map_structure(lambda ys__, y_: ys__.push_back(y_), ys_, y)
+            ys_ = tree.map_structure(lambda ys__, y_: ys__.push_back(y_) if ys__ is not None else None, ys_, y)
             return i + 1, s, ys_
 
         _, final_s, ys = while_loop(
@@ -197,12 +197,12 @@ def scan(
             (
                 rf.constant(0, dtype=rf.get_default_array_index_dtype(), dims=()),  # i
                 initial,  # state
-                tree.map_structure(lambda y: TensorArray(y), ys),
+                tree.map_structure(lambda y: TensorArray(y) if y is not None else None, ys),
             ),
         )
 
     if not return_tensor_arrays:
-        ys = tree.map_structure(lambda ys_: ys_.stack(axis=spatial_dim), ys)
+        ys = tree.map_structure(lambda ys_: ys_.stack(axis=spatial_dim) if ys_ is not None else None, ys)
     return ys, final_s, spatial_dim
 
 

{returnn-1.20230816.14926 → returnn-1.20230816.20915}/returnn/frontend/tensor_array.py

@@ -37,6 +37,7 @@ class TensorArray:
         _backend: Optional[Type[Backend]] = None,
         _enable_delayed_check: bool = False,
     ):
+        assert isinstance(tensor_template, Tensor)
         self.tensor_template = tensor_template
         if _backend is None:
             _backend = global_backend

{returnn-1.20230816.14926 → returnn-1.20230816.20915}/returnn/tf/frontend_layers/_backend.py

@@ -688,7 +688,6 @@ class ReturnnLayersBackend(Backend[Layer]):
         indices = []
         for i, a in enumerate(axes):
             assert isinstance(a, Dim)
-            assert not a.need_masking()  # not supported currently
             sub_name = "indices" if single_axis else f"indices{i}"
             indices_data = values.copy_template(name=f"{values.name}_{sub_name}_{a.description}")
             indices_data.dtype = "int32"

{returnn-1.20230816.14926 → returnn-1.20230816.20915/returnn.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: returnn
-Version: 1.20230816.14926
+Version: 1.20230816.20915
 Summary: The RWTH extensible training framework for universal recurrent neural networks
 Home-page: https://github.com/rwth-i6/returnn/
 Author: Albert Zeyer

returnn-1.20230816.14926/_setup_info_generated.py

@@ -1,2 +0,0 @@
-version = '1.20230816.014926'
-long_version = '1.20230816.014926+git.a54388f'