returnn 1.20250703.183400__py3-none-any.whl → 1.20250708.165746__py3-none-any.whl

returnn/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: returnn
-Version: 1.20250703.183400
+Version: 1.20250708.165746
 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 = '1.20250703.183400'
-long_version = '1.20250703.183400+git.a5c35a2'
+version = '1.20250708.165746'
+long_version = '1.20250708.165746+git.eff9959'

returnn/datasets/lm.py

@@ -9,6 +9,7 @@ from __future__ import annotations
 
 from typing import (
     Iterable,
+    Literal,
     Optional,
     Sequence,
     Union,
@@ -1177,6 +1178,7 @@ class PhoneSeqGenerator:
         add_extra_begin_lemma: float = 1.0,
         extra_end_lemma: Optional[Dict[str, Any]] = None,
         add_extra_end_lemma: float = 1.0,
+        phon_pick_strategy: Literal["random", "first"] = "random",
     ):
         """
         :param lexicon_file: lexicon XML file
@@ -1196,6 +1198,8 @@ class PhoneSeqGenerator:
         :param add_extra_begin_lemma:
         :param extra_end_lemma: just like ``extra_begin_lemma``, but for the end
         :param add_extra_end_lemma:
+        :param phon_pick_strategy: "random" or "first". If "random", then lemmas are picked randomly
+            if multiple pronunciations exist.
         """
         self.lexicon = Lexicon(lexicon_file)
         self.phonemes = sorted(self.lexicon.phonemes.keys(), key=lambda s: self.lexicon.phonemes[s]["index"])
@@ -1217,6 +1221,7 @@ class PhoneSeqGenerator:
         self.add_extra_begin_lemma = add_extra_begin_lemma
         self.extra_end_lemma = extra_end_lemma
         self.add_extra_end_lemma = add_extra_end_lemma
+        self.phon_pick_strategy = phon_pick_strategy
 
     def random_seed(self, seed: int):
         """Reset RNG via given seed"""
@@ -1284,7 +1289,12 @@ class PhoneSeqGenerator:
         """:return: space-separated phones"""
         phones = []
         for lemma in self._iter_orth_lemmas(orth):
-            phon = self.rnd.choice(lemma["phons"])
+            if self.phon_pick_strategy == "first":
+                phon = lemma["phons"][0]
+            elif self.phon_pick_strategy == "random":
+                phon = self.rnd.choice(lemma["phons"])
+            else:
+                raise ValueError(f"Unknown phon_pick_strategy {self.phon_pick_strategy}")
             phones.append(phon["phon"])
         return " ".join(phones)
 
@@ -1356,7 +1366,13 @@ class PhoneSeqGenerator:
         """
         allos: List[AllophoneState] = []
         for lemma in self._iter_orth_lemmas(orth):
-            phon = self.rnd.choice(lemma["phons"])  # space-separated phones in phon["phon"]
+            if self.phon_pick_strategy == "first":
+                phon = lemma["phons"][0]
+            elif self.phon_pick_strategy == "random":
+                phon = self.rnd.choice(lemma["phons"])
+            else:
+                raise ValueError(f"Unknown phon_pick_strategy {self.phon_pick_strategy}")
+            # space-separated phones in phon["phon"]
             l_allos = list(self._phones_to_allos(phon["phon"].split()))
             l_allos[0].mark_initial()
             l_allos[-1].mark_final()

returnn/datasets/meta.py

@@ -1990,9 +1990,13 @@ class VariableDataset(Dataset):
 class MultiEpochDataset(CachedDataset2):
     """
     It wraps some dataset, where one outer epoch corresponds to multiple epochs in the inner wrapped dataset.
+    I.e. one iteration through this dataset corresponds to multiple iterations through the inner dataset.
 
-    This can be useful when the inner dataset uses partition_epoch, and we want to cover the whole full epoch.
+    This can be useful for forwarding, when you want to do multiple iterations through the dataset.
+    This could be useful for clustering.
 
+    This can also be useful when the inner dataset uses (or must use) partition_epoch,
+    and we want to cover the whole full epoch:
     One specific example when the data is distributed over multiple files,
     and for reasonable performance, you want to have the data copied to the local disk,
     but all data together is too large to fit on the local disk.
@@ -2041,7 +2045,11 @@ class MultiEpochDataset(CachedDataset2):
         return self._dataset.get_all_tags()
 
     def get_total_num_seqs(self, *, fast: bool = False) -> int:
-        """total num seqs"""
+        """
+        Total num seqs.
+        Note that this is the total number of seqs in the inner dataset,
+        so without the multi-epoch handling.
+        """
         return self._dataset.get_total_num_seqs(fast=fast)
 
     def get_data_keys(self) -> List[str]:

{returnn-1.20250703.183400.dist-info → returnn-1.20250708.165746.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: returnn
-Version: 1.20250703.183400
+Version: 1.20250708.165746
 Summary: The RWTH extensible training framework for universal recurrent neural networks
 Home-page: https://github.com/rwth-i6/returnn/
 Author: Albert Zeyer

{returnn-1.20250703.183400.dist-info → returnn-1.20250708.165746.dist-info}/RECORD

@@ -1,9 +1,9 @@
-returnn/PKG-INFO,sha256=6b050edAnN_IixiTU7cKY9Y7GsV2u54Go4pMebtP1LA,5215
+returnn/PKG-INFO,sha256=gFz3WCtUJq8MCZeykNbFyHQVIFyQnwVcQkfRsDeC0X0,5215
 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=W3RqBPspEAo2psz8RlzIp3A0pBAmerlhlAfLeZuEUy8,77
+returnn/_setup_info_generated.py,sha256=QIOgjEYWfaIUKGZdpx_bvAWra8S_SM7m87LEcq2Lcp0,77
 returnn/config.py,sha256=3tmKhB6FnQZaNdtcYsiB61JnEY--iZ2qmJ4yq0b6tE0,29140
 returnn/forward_iface.py,sha256=A_OJiaXsX4MlXQRzST86ylyxSUZbC402PQL1REcqHjM,911
 returnn/learning_rate_control.py,sha256=ZvWryAn_tv9DhV8sh1LV3eE34Yltl3On3mYZAG4hR9s,34684
@@ -20,9 +20,9 @@ returnn/datasets/cached2.py,sha256=_6pza3IG68JexaExhj1ld3fP6pE7T-G804driJ9Z_qo,1
 returnn/datasets/distrib_files.py,sha256=9-3pJaF8Ws1Cs4AlelFCODz6b5YiaTsrD7tMCB76PDY,29865
 returnn/datasets/generating.py,sha256=9U_w6URIrv-Rb-hDbPOzYW9qYXzJbw32N6G268IKyoM,99833
 returnn/datasets/hdf.py,sha256=v5sjBenURR9Z-g7AQ9tsL84yDSye5RtbLpym3M6HSDE,67833
-returnn/datasets/lm.py,sha256=IqUsOzbdSWUynL0YFL25HbtMR4AxaQGHvjjqRE9IwBo,99215
+returnn/datasets/lm.py,sha256=ycHdGHxT4QshBM9LPktLDaaQRTLO5zQyueCK5KMNR_4,100022
 returnn/datasets/map.py,sha256=kOBJVZmwDhLsOplzDNByIfa0NRSUaMo2Lsy36lBvxrM,10907
-returnn/datasets/meta.py,sha256=KQtidTgSh-1gNgbpJ8OhXt6v2lkhPPH5dpjfzwsr3E4,95251
+returnn/datasets/meta.py,sha256=6XPPxhiNSxWw9Hu5Z6wG8dD9Zk82FqiI-k9HGQSTKgw,95658
 returnn/datasets/multi_proc.py,sha256=aVjsLt2qjHnHOrEYCgIPCwNYE-f1fiGP6eZ8NGAr3A4,22583
 returnn/datasets/normalization_data.py,sha256=J3njQCMvWAbIAVPepO2L_Xdau9eWYB7Zyd6STeGzTbc,14615
 returnn/datasets/numpy_dump.py,sha256=wl8bKIKAlff2HPJPtuu5wBg3TLOf16d2wLVB4lLAwTM,5158
@@ -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-1.20250703.183400.dist-info/LICENSE,sha256=ywBD_U2aD4vpuoIgNAsjIGBYydl0tVKll3De0Z8s77c,11041
-returnn-1.20250703.183400.dist-info/METADATA,sha256=6b050edAnN_IixiTU7cKY9Y7GsV2u54Go4pMebtP1LA,5215
-returnn-1.20250703.183400.dist-info/WHEEL,sha256=iAkIy5fosb7FzIOwONchHf19Qu7_1wCWyFNR5gu9nU0,91
-returnn-1.20250703.183400.dist-info/top_level.txt,sha256=Lsn4WZc5Pbfk0-xDQOgnFCxOoqxL4CyeM3N1TFbJncw,8
-returnn-1.20250703.183400.dist-info/RECORD,,
+returnn-1.20250708.165746.dist-info/LICENSE,sha256=ywBD_U2aD4vpuoIgNAsjIGBYydl0tVKll3De0Z8s77c,11041
+returnn-1.20250708.165746.dist-info/METADATA,sha256=gFz3WCtUJq8MCZeykNbFyHQVIFyQnwVcQkfRsDeC0X0,5215
+returnn-1.20250708.165746.dist-info/WHEEL,sha256=iAkIy5fosb7FzIOwONchHf19Qu7_1wCWyFNR5gu9nU0,91
+returnn-1.20250708.165746.dist-info/top_level.txt,sha256=Lsn4WZc5Pbfk0-xDQOgnFCxOoqxL4CyeM3N1TFbJncw,8
+returnn-1.20250708.165746.dist-info/RECORD,,