returnn 1.20250719.212120__py3-none-any.whl → 1.20250724.195711__py3-none-any.whl

returnn/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: returnn
-Version: 1.20250719.212120
+Version: 1.20250724.195711
 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 = '1.20250719.212120'
-long_version = '1.20250719.212120+git.dd75ba0'
+version = '1.20250724.195711'
+long_version = '1.20250724.195711+git.e0cf62f'

returnn/datasets/basic.py

@@ -19,7 +19,6 @@ import os
 import math
 import numpy
 import functools
-import typing
 from typing import TYPE_CHECKING, Optional, Any, Set, Tuple, Union, Type, Dict, Sequence, List, Callable
 
 from returnn.log import log
@@ -428,9 +427,9 @@ class Dataset:
         """
         raise OptionalNotImplementedError
 
-    def get_num_timesteps(self):
+    def get_num_timesteps(self) -> Union[int, NumbersDict]:
         """
-        :rtype: int
+        :return: how much frames we have in total.
         """
         assert self._num_timesteps > 0
         return self._num_timesteps
@@ -559,7 +558,7 @@ class Dataset:
             for i in range(1, num):
                 seq_index[i::num] += i * (num_seqs // num)
         elif seq_ordering_method == "reverse":
-            seq_index = range(num_seqs - 1, -1, -1)  # type: Union[range, typing.Sequence[int]]
+            seq_index = range(num_seqs - 1, -1, -1)  # type: Union[range, Sequence[int]]
         elif seq_ordering_method in ["sorted", "sorted_reverse"]:
             assert get_seq_len
             reverse = -1 if seq_ordering_method == "sorted_reverse" else 1
@@ -748,12 +747,11 @@ class Dataset:
         """
         self.epoch = None
 
-    def get_current_seq_order(self):
+    def get_current_seq_order(self) -> Sequence[int]:
         """
         :return: many datasets use self.get_seq_order_for_epoch. this function would return the current seq order
           for the current epoch, after self.init_seq_order was called.
           Not all datasets implement this.
-        :rtype: typing.Sequence[int]
         """
         raise OptionalNotImplementedError
 
@@ -902,7 +900,7 @@ class Dataset:
         if self.seq_ordering == "default" and self.partition_epoch == 1:
             return seq_idx
         assert self.have_corpus_seq_idx()
-        raise NotImplemented
+        raise NotImplementedError
 
     def have_get_corpus_seq(self) -> bool:
         """
@@ -1061,7 +1059,7 @@ class Dataset:
         if key in self.num_outputs:
             if self.num_outputs[key][1] <= 1:
                 return []
-            res_shape = [None] * (self.num_outputs[key][1] - 1)  # type: typing.List[typing.Union[None,int]]
+            res_shape: List[Union[None, int]] = [None] * (self.num_outputs[key][1] - 1)
             if not self.is_data_sparse(key):
                 res_shape[-1] = self.get_data_dim(key)
             return res_shape

returnn/datasets/cached2.py

@@ -4,18 +4,10 @@ Provides :class:`CachedDataset2`.
 
 from __future__ import annotations
 import numpy
-import typing
-from typing import Optional
+from typing import Optional, List
 from threading import Condition
 from .basic import Dataset, DatasetSeq
 
-try:
-    # noinspection PyCompatibility
-    from _thread import interrupt_main
-except ImportError:
-    # noinspection PyUnresolvedReferences,PyCompatibility
-    from thread import interrupt_main
-
 
 class CachedDataset2(Dataset):
     """
@@ -36,7 +28,7 @@ class CachedDataset2(Dataset):
         self._num_timesteps = None
         self.epoch = None
         self.reached_final_seq = False
-        self.added_data = []  # type: typing.List[DatasetSeq]
+        self.added_data: List[DatasetSeq] = []
         self.expected_load_seq_start = 0
         self._num_timesteps_accumulated = 0
 

returnn/datasets/lm.py

@@ -24,7 +24,6 @@ from typing import (
     cast,
     Generator,
 )
-import typing
 import os
 from io import IOBase
 import sys
@@ -1563,7 +1562,6 @@ class TranslationDataset(CachedDataset2):
             import returnn.util.better_exchook
 
             returnn.util.better_exchook.install()
-            from returnn.util.basic import AsyncThreadRun
 
             # First iterate once over the data to get the data len as fast as possible.
             data_len = 0

returnn/frontend/decoder/transformer.py

@@ -268,6 +268,7 @@ class TransformerDecoderLayer(rf.Module):
         ] = None,
         self_att_opts: Optional[Dict[str, Any]] = None,
         att_dropout: float = 0.1,
+        cross_att: Optional[Dict[str, Any]] = None,
         norm: Union[type, Dict[str, Any], rf.Module, Callable] = rf.LayerNorm,
     ):
         """
@@ -333,10 +334,10 @@ class TransformerDecoderLayer(rf.Module):
             raise TypeError(f"unexpected self_att type {self_att!r}")
         self.self_att_layer_norm = make_norm(norm, out_dim)
 
-        self.cross_att = None
+        self.cross_att: Optional[rf.CrossAttention] = None  # type might be inaccurate, but we expect this interface
         self.cross_att_layer_norm = None
         if encoder_dim is not None:
-            self.cross_att = rf.CrossAttention(
+            cross_att_opts = dict(
                 encoder_dim=self.encoder_dim,
                 query_in_dim=out_dim,
                 proj_dim=out_dim,
@@ -345,6 +346,12 @@ class TransformerDecoderLayer(rf.Module):
                 num_heads=num_heads,
                 att_dropout=att_dropout,
             )
+            if cross_att is None:
+                self.cross_att = rf.CrossAttention(**cross_att_opts)
+            elif isinstance(cross_att, dict):
+                self.cross_att: Optional[rf.CrossAttention] = rf.build_from_dict(cross_att, **cross_att_opts)
+            else:
+                raise TypeError(f"unexpected cross_att type {cross_att!r}")
             self.cross_att_layer_norm = make_norm(norm, out_dim)
 
     def default_initial_state(self, *, batch_dims: Sequence[Dim]) -> rf.State:

{returnn-1.20250719.212120.dist-info → returnn-1.20250724.195711.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: returnn
-Version: 1.20250719.212120
+Version: 1.20250724.195711
 Summary: The RWTH extensible training framework for universal recurrent neural networks
 Home-page: https://github.com/rwth-i6/returnn/
 Author: Albert Zeyer

{returnn-1.20250719.212120.dist-info → returnn-1.20250724.195711.dist-info}/RECORD

@@ -1,9 +1,9 @@
-returnn/PKG-INFO,sha256=kFAM5n3kH99yMVP1SQd0sVbkU2B7US9b0vzffW_Qd1U,5215
+returnn/PKG-INFO,sha256=fdsXa-aGlgSS206_rjbOuzyZOYFra39xWql2ejHX1ic,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=rBsXXneRSaGVcWW-gkfggWzjHOl--i-R6QX1-LCZ3r4,77
+returnn/_setup_info_generated.py,sha256=xJWrWKUD6b2uMW_UOzZMqg4UjzzWUtDTGUnDvJiHAOI,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
@@ -13,14 +13,14 @@ returnn/native_op.py,sha256=4_NnvfNxsM8GE_FsD6yOg6PZegqIdtJ3Sl1GdBWmFvg,244424
 returnn/pretrain.py,sha256=MHiXJZqkQFmDVyaYsGpd_Acv20wxl7Pr6s6qJzAT2FI,22648
 returnn/datasets/__init__.py,sha256=PvDlfDOaaopIeUIt0OSvHD2eHZkdkyE-sjMXf35EH5U,390
 returnn/datasets/audio.py,sha256=Gmj7a08dnvYh7Z-G1TNapz42L50AIcDE9JeIZaO1s1M,23334
-returnn/datasets/basic.py,sha256=S7MoFnQfgjeuZsBnaPrpQWlMUmH68HlUnjX7k881PD0,72302
+returnn/datasets/basic.py,sha256=IJhytVPiQZi7BD8-JVziKKT__PE528FwLQmbeiVQHzc,72303
 returnn/datasets/bundle_file.py,sha256=KQNrS1MSf-4_idlK0c0KFwON-f5sEK0sWU15WpoMYpE,2380
 returnn/datasets/cached.py,sha256=RyefRjSDdp-HveK-2vLy2C6BIHcpqQ_lNvUKlIa4QAI,25412
-returnn/datasets/cached2.py,sha256=_6pza3IG68JexaExhj1ld3fP6pE7T-G804driJ9Z_qo,12141
+returnn/datasets/cached2.py,sha256=oJOq2lWRQpxm6kyUKW1w5qZBd4kdKEpwM7KY_QnXbq4,11922
 returnn/datasets/distrib_files.py,sha256=SJ2YkZEZmG9lu3MLTwSMyVNfsXzRHqbLNjUn9IDwVJM,30194
 returnn/datasets/generating.py,sha256=9U_w6URIrv-Rb-hDbPOzYW9qYXzJbw32N6G268IKyoM,99833
 returnn/datasets/hdf.py,sha256=v5sjBenURR9Z-g7AQ9tsL84yDSye5RtbLpym3M6HSDE,67833
-returnn/datasets/lm.py,sha256=ycHdGHxT4QshBM9LPktLDaaQRTLO5zQyueCK5KMNR_4,100022
+returnn/datasets/lm.py,sha256=rQ3jV43lSnlGkKu7m5jTTH7aK0BOMXQocsHfJ8OGec8,99950
 returnn/datasets/map.py,sha256=kOBJVZmwDhLsOplzDNByIfa0NRSUaMo2Lsy36lBvxrM,10907
 returnn/datasets/meta.py,sha256=6XPPxhiNSxWw9Hu5Z6wG8dD9Zk82FqiI-k9HGQSTKgw,95658
 returnn/datasets/multi_proc.py,sha256=aVjsLt2qjHnHOrEYCgIPCwNYE-f1fiGP6eZ8NGAr3A4,22583
@@ -135,7 +135,7 @@ returnn/frontend/conversions/espnet_e_branchformer.py,sha256=Mmp3G6nySy0CqeHa-um
 returnn/frontend/conversions/hf_llama.py,sha256=1WQOhQyUWwkAznaRqK2zpThP8XZbaomkaE8qMG_bZPY,9662
 returnn/frontend/conversions/torch_nn.py,sha256=WAq_hs1tb5OC4iGmVemXvo3qba_e1MJXxRzG9pNK2HI,2204
 returnn/frontend/decoder/__init__.py,sha256=A-koKyPVlXp_V_2bk6GKZ1Xfv4rYIcfxGMXQHkHZiOQ,41
-returnn/frontend/decoder/transformer.py,sha256=GzZ-1HxDjmzM8bsW9KZn-4_W97vsnChbo_NtL-OBYJA,23179
+returnn/frontend/decoder/transformer.py,sha256=20a37hMiPbQBHx3tSbOeiAbFPVRcX_KYpPuw8tmY6GU,23658
 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=ro0uzEzDbAyNGYN5ff0KmiDl4HOYQluu64mJxYzuy-M,19972
@@ -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-1.20250719.212120.dist-info/LICENSE,sha256=ywBD_U2aD4vpuoIgNAsjIGBYydl0tVKll3De0Z8s77c,11041
-returnn-1.20250719.212120.dist-info/METADATA,sha256=kFAM5n3kH99yMVP1SQd0sVbkU2B7US9b0vzffW_Qd1U,5215
-returnn-1.20250719.212120.dist-info/WHEEL,sha256=iAkIy5fosb7FzIOwONchHf19Qu7_1wCWyFNR5gu9nU0,91
-returnn-1.20250719.212120.dist-info/top_level.txt,sha256=Lsn4WZc5Pbfk0-xDQOgnFCxOoqxL4CyeM3N1TFbJncw,8
-returnn-1.20250719.212120.dist-info/RECORD,,
+returnn-1.20250724.195711.dist-info/LICENSE,sha256=ywBD_U2aD4vpuoIgNAsjIGBYydl0tVKll3De0Z8s77c,11041
+returnn-1.20250724.195711.dist-info/METADATA,sha256=fdsXa-aGlgSS206_rjbOuzyZOYFra39xWql2ejHX1ic,5215
+returnn-1.20250724.195711.dist-info/WHEEL,sha256=iAkIy5fosb7FzIOwONchHf19Qu7_1wCWyFNR5gu9nU0,91
+returnn-1.20250724.195711.dist-info/top_level.txt,sha256=Lsn4WZc5Pbfk0-xDQOgnFCxOoqxL4CyeM3N1TFbJncw,8
+returnn-1.20250724.195711.dist-info/RECORD,,