returnn 1.20250211.110723__py3-none-any.whl → 1.20250211.215623__py3-none-any.whl

returnn/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: returnn
-Version: 1.20250211.110723
+Version: 1.20250211.215623
 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 = '1.20250211.110723'
-long_version = '1.20250211.110723+git.f032108'
+version = '1.20250211.215623'
+long_version = '1.20250211.215623+git.8640885'

returnn/datasets/lm.py

@@ -1087,8 +1087,17 @@ class Lexicon:
                             {"phon": e.text.strip(), "score": float(e.attrib.get("score", 0))}
                             for e in elem.findall("phon")
                         ]
-                        assert orth not in self.lemmas
-                        self.lemmas[orth] = {"orth": orth, "phons": phons}
+                        lemma = {"orth": orth, "phons": phons}
+                        if orth in self.lemmas:  # unexpected, already exists?
+                            if self.lemmas[orth] == lemma:
+                                print(f"Warning: lemma {lemma} duplicated in lexicon {filename}", file=log.v4)
+                            else:
+                                raise Exception(
+                                    f"orth {orth!r} lemma duplicated in lexicon {filename}."
+                                    f" old: {self.lemmas[orth]}, new: {lemma}"
+                                )
+                        else:  # lemma does not exist yet -- this is the expected case
+                            self.lemmas[orth] = lemma
                     root.clear()  # free memory
         print("Finished whole lexicon, %i lemmas" % len(self.lemmas), file=log.v4)
 

returnn/frontend/_cache.py

@@ -39,6 +39,9 @@ class Cache:
         # not shared over all instances of this class.
         self._lru_cache = lru_cache(max_size)(_lru_cache_dummy_func)
 
+    def __repr__(self):
+        return f"<{self.__class__.__module__}.{self.__class__.__qualname__} {self._lru_cache.cache_info()}>"
+
     def get(self, key, default=None):
         """
         :param key:

returnn/torch/engine.py

@@ -980,6 +980,7 @@ class Engine(EngineBase):
                 missing_keys_preload, unexpected_keys_preload = self._pt_model.load_state_dict(
                     preload_model_state, strict=False
                 )
+                preload_model_state_keys = set(preload_model_state.keys())
                 loaded_state_keys.update(preload_model_state.keys())
                 missing_keys.difference_update(preload_model_state.keys())
                 del preload_model_state
@@ -987,6 +988,11 @@ class Engine(EngineBase):
 
                 if opts.get("prefix", ""):
                     prefix_keys = [key for key in self._pt_model.state_dict() if key.startswith(opts.get("prefix", ""))]
+                    if not prefix_keys:
+                        raise Exception(
+                            "No keys with prefix %r found in model.\nModel params:\n%s"
+                            % (opts.get("prefix", ""), ", ".join(name for name, _ in self._pt_model.named_parameters()))
+                        )
                 else:
                     prefix_keys = model_state_keys_set
                 missing_keys_preload = (
@@ -995,6 +1001,12 @@ class Engine(EngineBase):
                 unexpected_keys_preload = (
                     set(prefix_keys).intersection(set(unexpected_keys_preload)).difference(loaded_state_keys)
                 )
+                if not preload_model_state_keys.intersection(prefix_keys):
+                    raise Exception(
+                        f"No keys with prefix {opts.get('prefix', '')!r} found in preload model state.\n"
+                        f"Preload model state keys: {preload_model_state_keys}\n"
+                        f"Model state keys: {model_state_keys_set}"
+                    )
                 if missing_keys_preload and not opts.get("ignore_missing", False):
                     missing_keys.update(missing_keys_preload)
                 if missing_keys_preload:

returnn/torch/frontend/bridge.py

@@ -109,18 +109,27 @@ class RFModuleAsPTModule(torch.nn.Module):
         self._aux_params_as_buffers = aux_params_as_buffers
         self._is_initializing = True
 
-        # recurse=False because param names cannot contain ".", will add submodules below recursively.
-        for name, rf_param in rf_module.named_parameters(recurse=False):
-            pt_param = rf_param.raw_tensor
-            assert isinstance(pt_param, torch.nn.Parameter)
-            if rf_param.auxiliary and aux_params_as_buffers:
-                self.register_buffer(name, pt_param)
-            else:
-                self.register_parameter(name, pt_param)
-
-        for name, rf_mod in rf_module.named_children():
-            pt_mod = rf_module_to_pt_module(rf_mod, aux_params_as_buffers=aux_params_as_buffers)
-            self.add_module(name, pt_mod)
+        for name, value in vars(rf_module).items():
+            if isinstance(value, rf.Parameter):
+                pt_param = value.raw_tensor
+                assert isinstance(pt_param, torch.nn.Parameter)
+                if value.auxiliary and aux_params_as_buffers:
+                    self.register_buffer(name, pt_param)
+                else:
+                    self.register_parameter(name, pt_param)
+
+            elif isinstance(value, rf.Module):
+                pt_mod = rf_module_to_pt_module(value, aux_params_as_buffers=aux_params_as_buffers)
+                self.add_module(name, pt_mod)
+
+            elif isinstance(value, torch.nn.Parameter):
+                self.register_parameter(name, value)
+
+            elif isinstance(value, torch.Tensor):  # make sure this check is after torch.nn.Parameter
+                self.register_buffer(name, value)
+
+            elif isinstance(value, torch.nn.Module):
+                self.add_module(name, value)
 
         self._is_initializing = False
 

returnn/util/lru_cache.py

@@ -196,7 +196,7 @@ def _lru_cache_wrapper(user_function, maxsize: int, typed: bool):
         """
         Removes the entry from the cache.
         """
-        nonlocal hits, misses
+        nonlocal hits, misses, full
         key = make_key(args, kwargs, typed)
         with lock:
             link = cache_get(key)
@@ -208,6 +208,7 @@ def _lru_cache_wrapper(user_function, maxsize: int, typed: bool):
                 oldvalue = link[RESULT]
                 link.clear()
                 del cache[oldkey]
+                full = cache_len() >= maxsize
                 return oldvalue
             if fallback is not_specified:
                 raise KeyError("key not found")

{returnn-1.20250211.110723.dist-info → returnn-1.20250211.215623.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: returnn
-Version: 1.20250211.110723
+Version: 1.20250211.215623
 Summary: The RWTH extensible training framework for universal recurrent neural networks
 Home-page: https://github.com/rwth-i6/returnn/
 Author: Albert Zeyer

{returnn-1.20250211.110723.dist-info → returnn-1.20250211.215623.dist-info}/RECORD

@@ -1,9 +1,9 @@
-returnn/PKG-INFO,sha256=URxQfD6qddQqyGuoM5_5h2XS9kTA0Qv6EZDqCCD3Pxk,5215
+returnn/PKG-INFO,sha256=1no-Aukog5H0Mvgs1485nTteX29Fjp-sH7_kGvbMHG4,5215
 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=R-VEKbgSRUuCxbz9CHRW-XZyUzwY_q_kEm5AZAMfMpM,77
+returnn/_setup_info_generated.py,sha256=xvdH59j45tyParGsCGfKR4RXzo9TQAImKpm20b4ylGk,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
@@ -20,7 +20,7 @@ returnn/datasets/cached2.py,sha256=STojLL2Ivvd0xMfZRlYgzsHKlikYKL-caZCIDCgc_9g,1
 returnn/datasets/distrib_files.py,sha256=kyqIQILDPAO2TXr39hjslmDxIAc3pkY1UOoj8nuiFXo,27534
 returnn/datasets/generating.py,sha256=e2-SXcax7xQ4fkVW_Q5MgOLP6KlB7EQXJi_v64gVAWI,99805
 returnn/datasets/hdf.py,sha256=shif0aQqWWNJ0b6YnycpPjIVNsxjLrA41Y66-_SluGI,66993
-returnn/datasets/lm.py,sha256=CX06bc06hAvf3c9-Gku_2DmqexgmR9TlRxsSPzRPno4,98193
+returnn/datasets/lm.py,sha256=h0IHUbze87njKrcD5eT1FRxde7elIio05n-BWiqmjFE,98805
 returnn/datasets/map.py,sha256=kOBJVZmwDhLsOplzDNByIfa0NRSUaMo2Lsy36lBvxrM,10907
 returnn/datasets/meta.py,sha256=wHquywF1C7-YWhcSFSAdDNc0nEHRjE-ks7YIEuDFMIE,94731
 returnn/datasets/multi_proc.py,sha256=7kppiXGiel824HM3GvHegluIxtiNAHafm-e6qh6W7YU,21948
@@ -76,7 +76,7 @@ returnn/extern/graph_editor/transform.py,sha256=d9fEgu0JC342q0g9niVxRWMKzkQQA9mr
 returnn/extern/graph_editor/util.py,sha256=QMrQeQZ7lJwsrNQub9tof0h3quEaoHiGJaZmogQ7jXE,18707
 returnn/frontend/__init__.py,sha256=2aS7nbxXniIrBp2DODl0xN0f3IJ_dX4Bi9ZlR7W5_DE,1472
 returnn/frontend/_backend.py,sha256=datTTGM6xhvpiSPphDZgRjruU-SxrowG15_ZnTkEKy0,50108
-returnn/frontend/_cache.py,sha256=f37Fayj90JeQKG8Kvmog73asDnzPRWK5yfDQ-DA0gGM,8269
+returnn/frontend/_cache.py,sha256=JAhi7L-raQ3A-NC3JUYDtdRTwT3BGJJGGZxrZ8MfEWQ,8403
 returnn/frontend/_numpy_backend.py,sha256=_akZZkZplzR8XNVIrenvR99s-CapWDYL9WSD7VXwGYI,7790
 returnn/frontend/_random_journal.py,sha256=_ktP_mjgx8vtQQGX_DofdhewJj0aPiczefTWeemPkmo,5457
 returnn/frontend/_utils.py,sha256=4A3MSRM0i86J77550uR_AjcBEPu6nymLUZ9Xd1V3Fkc,12073
@@ -207,7 +207,7 @@ returnn/tf/util/open_fst.py,sha256=sZRDw4TbxvhGqpGdUJWy1ebvlZm4_RPhygpRw9uLAOQ,1
 returnn/torch/README.md,sha256=jzJ2FpOHW02vxN69yKaV97C9LI-hmvjBglKfdZXIDdc,85
 returnn/torch/__init__.py,sha256=MHEUyNHB20Vy89uKAqZoj6FxJKF1Gq3HW-i6ra1pNcI,24
 returnn/torch/distributed.py,sha256=i13cUVjI7GxpO0TAresrNyCM0ZBAaf-cXNr09Fmg_2k,6266
-returnn/torch/engine.py,sha256=BjPF6Kifu1b8kCavnCGAXrp0uhqY4WFexLvP3kR6tmg,76322
+returnn/torch/engine.py,sha256=8BIpdcrpbJL9HrvCX-hISh-14zW9aSrHGvRWT9s0zOk,77103
 returnn/torch/updater.py,sha256=GqtBvZpElPVMm0lq84JPl4NVLFFETZAzAbR0rTomSao,28249
 returnn/torch/data/__init__.py,sha256=6cLNEi8KoGI12PF6akN7mI_mtjlx-0hcQAfMYoExwik,132
 returnn/torch/data/extern_data.py,sha256=_uT_9_gd5HIh1IoRsrebVG-nufSnb7fgC5jyU05GxJg,7580
@@ -218,7 +218,7 @@ returnn/torch/data/tensor_utils.py,sha256=-Teqi--LLbt6q_5mDRdoHZHmPgSdC83W706uki
 returnn/torch/frontend/__init__.py,sha256=AA48HZnC17ASuKA0EWy8loZ-Bib_yUtqF4T1wYvjst4,62
 returnn/torch/frontend/_backend.py,sha256=h_rUhBPxLRgpZSqX4C8vX8q4dHWMhZpwPmGbKN6MsZo,99995
 returnn/torch/frontend/_rand.py,sha256=1JgIkV2XmpgJD86zXZ-NCAe-QuoP2swr6NaS1oz3Qa8,1830
-returnn/torch/frontend/bridge.py,sha256=bAzOVlL-3hD6af9ir8EOyBSXy6O3KtnCRD7SaZTF2yU,7538
+returnn/torch/frontend/bridge.py,sha256=Z2_UW8AagezC7zsXDc5PKcd8G9WwisV7j9SWGHU0m4U,7840
 returnn/torch/frontend/raw_ops.py,sha256=lF0h-KtYYsdaaqQADylVZp9qzPskOOXA4MfmYDyx5IU,296
 returnn/torch/optim/README.md,sha256=0iH5FiKb7iDrVK5n8V6yCh4ciCFG2YSbyh7lPneT5ik,360
 returnn/torch/optim/__init__.py,sha256=yxdbnOkXAHzZ_t6cHi6zn5x_DQNlLZJ-KxZByHTIg1U,29
@@ -241,7 +241,7 @@ returnn/util/debug_helpers.py,sha256=0EINLK4uLtoSt5_kHs1M2NIFpMd0S7i4c4rx90U4fJk
 returnn/util/file_cache.py,sha256=JvJ4C7NFr8WpiIN0hLk3c33oX4-JfWSpchTjY7JGpCc,23127
 returnn/util/fsa.py,sha256=k2lJ8tyf_g44Xk1EPVLwDwpP4spoMTqIigDVOWocQHY,59177
 returnn/util/literal_py_to_pickle.py,sha256=3dnjWPeeiDT2xp4bRDgIf9yddx7b1AG7mOKEn_jiSl8,2173
-returnn/util/lru_cache.py,sha256=rQCImYpZ26PQGQ--GUzWzLfadDLfYEJv9vKmwwDTVJY,11391
+returnn/util/lru_cache.py,sha256=Kd6j_SitMERTfXvNKGdtzErKXkZ4jCJNETtaCv4HzKM,11443
 returnn/util/math.py,sha256=ximPqNsv0Wu6VNcCLqNfsmSu1s-VPsAJYt5nEvFZVtY,6691
 returnn/util/multi_proc_non_daemonic_spawn.py,sha256=YCW3Gry0RJ9Dsc5bKfZ77Q06eLjq6winGniYllJE7PU,9057
 returnn/util/native_code_compiler.py,sha256=T6eZwzNA7AnkpNpo61AbYVNVtKqdWBbQfvJEQVe-cHE,13172
@@ -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-1.20250211.110723.dist-info/LICENSE,sha256=ywBD_U2aD4vpuoIgNAsjIGBYydl0tVKll3De0Z8s77c,11041
-returnn-1.20250211.110723.dist-info/METADATA,sha256=URxQfD6qddQqyGuoM5_5h2XS9kTA0Qv6EZDqCCD3Pxk,5215
-returnn-1.20250211.110723.dist-info/WHEEL,sha256=P9jw-gEje8ByB7_hXoICnHtVCrEwMQh-630tKvQWehc,91
-returnn-1.20250211.110723.dist-info/top_level.txt,sha256=Lsn4WZc5Pbfk0-xDQOgnFCxOoqxL4CyeM3N1TFbJncw,8
-returnn-1.20250211.110723.dist-info/RECORD,,
+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,,