File size: 43,181 Bytes
d1ed09d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
import json
import logging
from abc import ABC, abstractmethod
from os import mkdir
from os.path import exists, isdir, isfile, join
from typing import Callable, Mapping, Optional, Sequence, Tuple

import torch


try:
    from safetensors.torch import load_file, save_file

    safetensors_available = True
except ImportError:
    safetensors_available = False

from .configuration import AdapterConfig, build_full_config
from .head_utils import STATIC_TO_FLEX_HEAD_MAP, get_head_config_and_rename_list
from .utils import (
    ACTIVATION_RENAME,
    ADAPTERFUSION_CONFIG_NAME,
    ADAPTERFUSION_WEIGHTS_NAME,
    CONFIG_NAME,
    HEAD_CONFIG_NAME,
    HEAD_WEIGHTS_NAME,
    SAFE_ADAPTERFUSION_WEIGHTS_NAME,
    SAFE_HEAD_WEIGHTS_NAME,
    SAFE_WEIGHTS_NAME,
    WEIGHTS_NAME,
    AdapterType,
    resolve_adapter_path,
)


logger = logging.getLogger(__name__)


class WeightsLoaderHelper:
    """
    A class providing helper methods for saving and loading module weights.
    """

    def __init__(
        self, model, weights_name, config_name, use_safetensors: bool = False, safe_weights_name: Optional[str] = None
    ):
        self.model = model
        self.weights_name = weights_name
        self.config_name = config_name
        self.use_safetensors = use_safetensors
        if use_safetensors and not safetensors_available:
            raise ValueError("Safetensors package not available. Please install via `pip install safetensors`.")
        self.safe_weights_name = safe_weights_name or weights_name

    def state_dict(self, filter_func):
        return {k: v for (k, v) in self.model.state_dict().items() if filter_func(k)}

    def rename_state_dict(self, state_dict, *rename_funcs):
        new_state_dict = {}
        for k, v in state_dict.items():
            new_k = k
            for rename_func in rename_funcs:
                new_k = rename_func(new_k)
            new_state_dict[new_k] = v
        return new_state_dict

    def save_weights_config(self, save_directory, config, meta_dict=None):
        # add meta information if given
        if meta_dict:
            for k, v in meta_dict.items():
                if k not in config:
                    config[k] = v
        # save to file system
        output_config_file = join(save_directory, self.config_name)
        with open(output_config_file, "w", encoding="utf-8") as f:
            json.dump(config, f, indent=2, sort_keys=True)
        logger.info("Configuration saved in {}".format(output_config_file))

    def save_weights(self, save_directory, filter_func):
        if not exists(save_directory):
            mkdir(save_directory)
        else:
            assert isdir(save_directory), "Saving path should be a directory where the module weights can be saved."

        # Get the state of all adapter modules for this task
        state_dict = self.state_dict(filter_func)
        # Save the adapter weights
        if self.use_safetensors:
            output_file = join(save_directory, self.safe_weights_name)
            save_file(state_dict, output_file)
        else:
            output_file = join(save_directory, self.weights_name)
            torch.save(state_dict, output_file)
        logger.info("Module weights saved in {}".format(output_file))

    def load_weights_config(self, save_directory):
        config_file = join(save_directory, self.config_name)
        logger.info("Loading module configuration from {}".format(config_file))
        # Load the config
        with open(config_file, "r", encoding="utf-8") as f:
            loaded_config = json.load(f)
        # For older versions translate the activation function to the new format
        if "version" not in loaded_config:
            if "config" in loaded_config and loaded_config["config"] is not None:
                if (
                    "non_linearity" in loaded_config["config"]
                    and loaded_config["config"]["non_linearity"] in ACTIVATION_RENAME
                ):
                    loaded_config["config"]["non_linearity"] = ACTIVATION_RENAME[
                        loaded_config["config"]["non_linearity"]
                    ]
        return loaded_config

    @staticmethod
    def _load_module_state_dict(module, state_dict, start_prefix=""):
        missing_keys = []
        unexpected_keys = []
        error_msgs = []

        # copy state_dict so _load_from_state_dict can modify it
        metadata = getattr(state_dict, "_metadata", None)
        state_dict = state_dict.copy()
        if metadata is not None:
            state_dict._metadata = metadata

        def load(module, prefix=""):
            local_metadata = {} if metadata is None else metadata.get(prefix[:-1], {})
            module._load_from_state_dict(
                state_dict, prefix, local_metadata, True, missing_keys, unexpected_keys, error_msgs
            )
            for name, child in module._modules.items():
                if child is not None:
                    load(child, prefix + name + ".")

        load(module, prefix=start_prefix)

        if len(error_msgs) > 0:
            raise RuntimeError(
                "Error(s) in loading state_dict for {}:\n\t{}".format(
                    module.__class__.__name__, "\n\t".join(error_msgs)
                )
            )
        return missing_keys, unexpected_keys

    def load_weights(
        self,
        save_directory,
        filter_func,
        rename_func=None,
        loading_info=None,
        in_base_model=False,
    ):
        # Load the weights of the adapter
        try:
            if self.use_safetensors:
                weights_file = join(save_directory, self.safe_weights_name)
                if exists(weights_file):
                    state_dict = load_file(weights_file, device="cpu")
                else:
                    logger.info(f"No safetensors file found in {save_directory}. Falling back to torch.load...")
                    weights_file = join(save_directory, self.weights_name)
                    state_dict = torch.load(weights_file, map_location="cpu")
            else:
                weights_file = join(save_directory, self.weights_name)
                state_dict = torch.load(weights_file, map_location="cpu")
        except Exception:
            raise OSError("Unable to load weights from pytorch checkpoint file. ")
        logger.info("Loading module weights from {}".format(weights_file))

        return self.load_weights_from_state_dict(
            state_dict, filter_func, rename_func=rename_func, loading_info=loading_info, in_base_model=in_base_model
        )

    def load_weights_from_state_dict(
        self, state_dict, filter_func, rename_func=None, loading_info=None, in_base_model=False, start_prefix=""
    ):
        # Rename weights if needed
        if rename_func:
            if isinstance(rename_func, Sequence):
                state_dict = self.rename_state_dict(state_dict, *rename_func)
            else:
                state_dict = self.rename_state_dict(state_dict, rename_func)

        # Add the weights to the model
        # Make sure we are able to load base models as well as derived models (with heads)
        model_to_load = self.model
        has_prefix_module = any(s.startswith(self.model.base_model_prefix) for s in state_dict.keys())
        if not start_prefix and not hasattr(self.model, self.model.base_model_prefix) and has_prefix_module:
            start_prefix = self.model.base_model_prefix + "."
        if in_base_model and hasattr(self.model, self.model.base_model_prefix) and not has_prefix_module:
            model_to_load = self.model.base_model

        missing_keys, unexpected_keys = self._load_module_state_dict(
            model_to_load, state_dict, start_prefix=start_prefix
        )

        missing_keys = [k for k in missing_keys if filter_func(k)]
        if len(missing_keys) > 0:
            logger.info(
                "Some module weights could not be found in loaded weights file: {}".format(", ".join(missing_keys))
            )
        if self.model._keys_to_ignore_on_load_unexpected:
            unexpected_keys = [k for k in unexpected_keys if k not in self.model._keys_to_ignore_on_load_unexpected]
        if len(unexpected_keys) > 0:
            logger.info(
                "Some weights of the state_dict could not be loaded into model: {}".format(", ".join(unexpected_keys))
            )

        if isinstance(loading_info, dict):
            if "missing_keys" not in loading_info:
                loading_info["missing_keys"] = []
            if "unexpected_keys" not in loading_info:
                loading_info["unexpected_keys"] = []
            loading_info["missing_keys"].extend(missing_keys)
            loading_info["unexpected_keys"].extend(unexpected_keys)

        return missing_keys, unexpected_keys


class WeightsLoader(ABC):
    """
    An abstract class providing basic methods for saving and loading weights of a model. Extend this class to build
    custom module weight loaders.
    """

    def __init__(
        self, model, weights_name, config_name, use_safetensors: bool = False, safe_weights_name: Optional[str] = None
    ):
        self.model = model
        self.weights_helper = WeightsLoaderHelper(
            model, weights_name, config_name, use_safetensors=use_safetensors, safe_weights_name=safe_weights_name
        )

    @abstractmethod
    def filter_func(self, name: str) -> Callable[[str], bool]:
        """
        The callable returned by this method is used to extract the module weights to be saved or loaded based on their
        names.

        Args:
            name (str): An identifier of the weights to be saved.

        Returns:
            Callable[str, bool]: A function that takes the fully qualified name of a module parameter and returns a
            boolean value that specifies whether this parameter should be extracted.
        """
        pass

    @abstractmethod
    def rename_func(self, old_name: str, new_name: str) -> Callable[[str], str]:
        """
        The callable returned by this method is used to optionally rename the module weights after loading.

        Args:
            old_name (str): The string identifier of the weights as loaded from file.
            new_name (str): The new string identifier to which the weights should be renamed.

        Returns:
            Callable[str, str]: A function that takes the fully qualified name of a module parameter and returns a new
            fully qualified name.
        """
        pass

    def save(self, save_directory, name, **kwargs):
        """
        Saves the module config and weights into the given directory. Override this method for additional saving
        actions.

        Args:
            save_directory (str): The directory to save the weights in.
            name (str): An identifier of the weights to be saved. The details are specified by the implementor.
        """
        if not exists(save_directory):
            mkdir(save_directory)
        else:
            assert isdir(
                save_directory
            ), "Saving path should be a directory where weights and configuration can be saved."

        config_dict = build_full_config(
            None,
            self.model.config,
            model_name=self.model.model_name,
            name=name,
            model_class=self.model.__class__.__name__,
        )
        meta_dict = kwargs.pop("meta_dict", None)

        # Save the adapter configuration
        self.weights_helper.save_weights_config(save_directory, config_dict, meta_dict=meta_dict)

        # Save adapter weights
        filter_func = self.filter_func(name)
        self.weights_helper.save_weights(save_directory, filter_func)

    def load(self, save_directory, load_as=None, loading_info=None, **kwargs) -> Tuple[str, str]:
        """
        Loads the module weights from the given directory. Override this method for additional loading actions. If
        adding the loaded weights to the model passed to the loader class requires adding additional modules, this
        method should also perform the architectural changes to the model.

        Args:
            save_directory (str): The directory from where to load the weights.
            load_as (str, optional): Load the weights with this name. Defaults to None.

        Returns:
            Tuple[str, str]: A tuple consisting of the local file system directory from which the weights where loaded
            and the name of the loaded weights.
        """
        if not exists(join(save_directory, self.weights_helper.weights_name)):
            raise ValueError("Loading path should be a directory where the weights are saved.")

        # Load config
        config = self.weights_helper.load_weights_config(save_directory)

        # Load head weights
        filter_func = self.filter_func(config["name"])
        if load_as:
            rename_func = self.rename_func(config["name"], load_as)
        else:
            rename_func = None
        self.weights_helper.load_weights(
            save_directory, filter_func, rename_func=rename_func, loading_info=loading_info
        )

        return save_directory, load_as or config["name"]


class AdapterLoader(WeightsLoader):
    """
    A class providing methods for saving and loading adapter modules from the Hub, the filesystem or a remote url.

    Model classes passed to this loader must implement the `ModelAdaptersMixin` class.
    """

    def __init__(self, model, adapter_type=None, use_safetensors: bool = False):
        super().__init__(
            model, WEIGHTS_NAME, CONFIG_NAME, use_safetensors=use_safetensors, safe_weights_name=SAFE_WEIGHTS_NAME
        )
        self.adapter_type = adapter_type
        if adapter_type and not AdapterType.has(self.adapter_type):
            raise ValueError("Invalid adapter type {}".format(self.adapter_type))

    def filter_func(self, adapter_name):
        return (
            lambda x: "_adapters.{}.".format(adapter_name) in x
            or ".adapters.{}.".format(adapter_name) in x
            or ".prefix_tunings.{}.".format(adapter_name) in x
            or ".prefix_gates.{}.".format(adapter_name) in x
            or ".loras.{}.".format(adapter_name) in x
            or ".refts.{}.".format(adapter_name) in x
            or ".prompt_tunings.{}.".format(adapter_name) in x
        )

    # This dict maps the original weight names to the currently used equivalents.
    # The mapping is used by rename_func() to support loading from older weights files.
    # Old adapters will be loaded and converted to the new format automatically.
    legacy_weights_mapping = {
        "attention_text_task_adapters": "adapters",
        "attention_text_lang_adapters": "adapters",
        "layer_text_task_adapters": "adapters",
        "layer_text_lang_adapters": "adapters",
        "invertible_lang_adapters": "invertible_adapters",
    }

    def _rename_legacy_weights(self, k):
        for old, new in self.legacy_weights_mapping.items():
            k = k.replace(old, new)
        return k

    # This method is used to remove unnecessary invertible adapters from task adapters using the old format.
    # In the old format, task adapters e.g. using seq_bn config specify inv. adapters but don't use them.
    # As inv. adapters would be incorrectly used in the new implementation,
    # catch this case here when loading pretrained adapters.
    def _fix_legacy_config(self, adapter_name, missing_keys):
        if self.adapter_type == AdapterType.text_task:
            inv_adapter_keys = [x for x in missing_keys if f"invertible_adapters.{adapter_name}." in x]
            if len(inv_adapter_keys) > 0:
                del self.model.base_model.invertible_adapters[adapter_name]
                missing_keys = [k for k in missing_keys if k not in inv_adapter_keys]
                # remove invertible_adapter from config
                adapter_config_name = self.model.adapters_config.adapters[adapter_name]
                if adapter_config_name in self.model.adapters_config.config_map:
                    adapter_config = self.model.adapters_config.config_map[adapter_config_name]
                    self.model.adapters_config.config_map[adapter_config_name] = adapter_config.replace(
                        inv_adapter=None, inv_adapter_reduction_factor=None
                    )
        return missing_keys

    def rename_func(self, old_name, new_name):
        return (
            lambda k: self._rename_legacy_weights(k)
            .replace("adapters.{}.".format(old_name), "adapters.{}.".format(new_name))
            .replace(".prefix_tunings.{}.".format(old_name), ".prefix_tunings.{}.".format(new_name))
            .replace(".prefix_gates.{}.".format(old_name), ".prefix_gates.{}.".format(new_name))
            .replace(".loras.{}.".format(old_name), ".loras.{}.".format(new_name))
            .replace(".refts.{}.".format(old_name), ".refts.{}.".format(new_name))
        )

    def save_to_state_dict(self, name: str):
        """
        Extracts the weights of a given adapter from the model and returns them as a state dict.

        Args:
            name (str): The name of the adapter to be saved.

        Returns:
            Tuple[dict, dict]: A tuple consisting of the state dict containing the adapter weights and the adapter
            configuration.
        """
        if name not in self.model.adapters_config.adapters:
            raise ValueError("No adapter of this type with the given name is part of this model.")

        adapter_config = self.model.adapters_config.get(name)

        config_dict = build_full_config(
            adapter_config,
            self.model.config,
            model_name=self.model.model_name,
            name=name,
            model_class=self.model.__class__.__name__,
        )

        # Save adapter weights
        filter_func = self.filter_func(config_dict["name"])
        state_dict = self.weights_helper.state_dict(filter_func)

        return state_dict, config_dict

    def save(self, save_directory, name, meta_dict=None):
        """
        Saves an adapter and its configuration file to a directory, so that it can be reloaded using the `load()`
        method.

        Args:
            save_directory (str): a path to a directory where the adapter will be saved
            task_name (str): the name of the adapter to be saved
        """
        if not exists(save_directory):
            mkdir(save_directory)
        else:
            assert isdir(
                save_directory
            ), "Saving path should be a directory where adapter and configuration can be saved."
        assert (
            name in self.model.adapters_config.adapters
        ), "No adapter of this type with the given name is part of this model."

        adapter_config = self.model.adapters_config.get(name)

        self.model.apply_to_adapter_layers(lambda _, layer: layer.pre_save_adapters())

        config_dict = build_full_config(
            adapter_config,
            self.model.config,
            model_name=self.model.model_name,
            name=name,
            model_class=self.model.__class__.__name__,
        )

        # Save the adapter configuration
        self.weights_helper.save_weights_config(save_directory, config_dict, meta_dict=meta_dict)

        # Save adapter weights
        filter_func = self.filter_func(config_dict["name"])
        self.weights_helper.save_weights(save_directory, filter_func)

    def load_from_state_dict(self, state_dict, name, load_as=None, loading_info=None, start_prefix=""):
        """
        Loads the weights of a given adapter from a state dict into the model.

        Args:
            state_dict (dict): The state dict from which to load the adapter weights.
            name (str): The name of the adapter to be loaded.
            load_as (str, optional): Load the adapter using this name. By default, the name with which the adapter was
                saved will be used.
            loading_info (dict, optional):
                A dictionary to which loading information (missing and unexpected keys) will be added.
            start_prefix (str, optional): A custom prefix to be ignored in the given state dict.
        """
        new_adapter_name = load_as or name
        if new_adapter_name not in self.model.adapters_config.adapters:
            raise ValueError("No adapter of this type with the given name is part of this model.")

        # Load adapter weights
        filter_func = self.filter_func(name)
        rename_func = self.rename_func(name, new_adapter_name)
        missing_keys, _ = self.weights_helper.load_weights_from_state_dict(
            state_dict,
            filter_func,
            rename_func=rename_func,
            loading_info=loading_info,
            in_base_model=True,
            start_prefix=start_prefix,
        )
        missing_keys = self._fix_legacy_config(new_adapter_name, missing_keys)
        if isinstance(loading_info, Mapping):
            loading_info["missing_keys"] = missing_keys

    def load(
        self,
        adapter_name_or_path,
        config=None,
        version=None,
        model_name=None,
        load_as=None,
        loading_info=None,
        leave_out=None,
        set_active=False,
        **kwargs,
    ):
        """
        Loads a pre-trained pytorch adapter module from the local file system or a remote location.

        Args:
            adapter_name_or_path (str): can be either:

                - the identifier of a pre-trained task adapter to be loaded from Adapter Hub
                - a path to a directory containing adapter weights saved using `model.saved_adapter()`
                - a URL pointing to a zip folder containing a saved adapter module
            config (str, optional): Deprecated.
            version (str, optional): The version of the adapter to be loaded.
            model_name (str, optional): Deprecated.
            load_as (str, optional): Load the adapter using this name. By default, the name with which the adapter was
             saved will be used.

        Returns:
            Tuple[str, str]: A tuple consisting of the local file system directory from which the weights where loaded
            and the name of the loaded weights.
        """
        # Warn about deprecated arguments
        if config is not None or model_name is not None:
            logger.warning(
                "The 'config' and 'model_name' arguments are specific to the now unsupported legacy Hub repo and will"
                " be removed."
                "Please switch to only providing the HF Model Hub identifier.",
            )
        requested_config = AdapterConfig.load(config) if config else None
        # Resolve the weights to be loaded based on the given identifier and the current adapter config
        model_name = self.model.model_name or model_name
        resolved_folder = resolve_adapter_path(
            adapter_name_or_path,
            model_name,
            adapter_config=requested_config,
            version=version,
            **kwargs,
        )

        # Load config of adapter
        config = self.weights_helper.load_weights_config(resolved_folder)
        if self.adapter_type and "type" in config:
            assert config["type"] == self.adapter_type, "Loaded adapter has to be a {} adapter.".format(
                self.adapter_type
            )
        elif "type" in config:
            self.adapter_type = config["type"]
        # post-loading drop of layers
        if leave_out is not None:
            if "leave_out" in config["config"] and config["config"]["leave_out"] is not None:
                # The conversion to a set and then back to a list removes all duplicates
                leave_out = list(set(leave_out + config["config"]["leave_out"]))
            config["config"]["leave_out"] = leave_out

        adapter_name = load_as or config["name"]
        # If the adapter is not part of the model, add it
        if adapter_name not in self.model.adapters_config.adapters:
            self.model.add_adapter(adapter_name, config=config["config"], set_active=set_active)
        else:
            logger.warning("Overwriting existing adapter '{}'.".format(adapter_name))

        # Load adapter weights
        filter_func = self.filter_func(adapter_name)
        rename_func = self.rename_func(config["name"], adapter_name)
        missing_keys, _ = self.weights_helper.load_weights(
            resolved_folder, filter_func, rename_func=rename_func, loading_info=loading_info, in_base_model=True
        )
        missing_keys = self._fix_legacy_config(adapter_name, missing_keys)
        if isinstance(loading_info, Mapping):
            loading_info["missing_keys"] = missing_keys

        return resolved_folder, adapter_name


class AdapterFusionLoader(WeightsLoader):
    """
    A class providing methods for saving and loading AdapterFusion modules from the file system.

    """

    def __init__(self, model, error_on_missing=True, use_safetensors: bool = False):
        super().__init__(
            model,
            ADAPTERFUSION_WEIGHTS_NAME,
            ADAPTERFUSION_CONFIG_NAME,
            use_safetensors=use_safetensors,
            safe_weights_name=SAFE_ADAPTERFUSION_WEIGHTS_NAME,
        )
        self.error_on_missing = error_on_missing

    def filter_func(self, adapter_fusion_name):
        return lambda x: "adapter_fusion_layer.{}".format(adapter_fusion_name) in x

    def rename_func(self, old_name, new_name):
        return lambda k: k.replace(
            "adapter_fusion_layer.{}".format(old_name), "adapter_fusion_layer.{}".format(new_name)
        )

    def save_to_state_dict(self, name: str):
        """
        Extracts the weights of a given AdapterFusion from the model and returns them as a state dict.

        Args:
            name (str): The name of the AdapterFusion to be saved.

        Returns:
            Tuple[dict, dict]: A tuple consisting of the state dict containing the AdapterFusion weights and the
            AdapterFusion configuration.
        """
        if name not in self.model.adapters_config.fusions:
            raise ValueError(f"No AdapterFusion with name '{name}' available.")

        adapter_fusion_config = self.model.adapters_config.get_fusion(name)

        config_dict = build_full_config(
            adapter_fusion_config,
            self.model.config,
            model_name=self.model.model_name,
            name=name,
            model_class=self.model.__class__.__name__,
        )

        # Save adapter weights
        filter_func = self.filter_func(name)
        state_dict = self.weights_helper.state_dict(filter_func)

        return state_dict, config_dict

    def save(self, save_directory: str, name: str, meta_dict=None):
        """
        Saves a AdapterFusion module into the given directory.

        Args:
            save_directory (str): The directory to save the weights in.
            name (str, optional): The AdapterFusion name.
        """

        if name not in self.model.adapters_config.fusions:
            if self.error_on_missing:
                raise ValueError(f"Unknown AdapterFusion '{name}'.")
            else:
                logger.debug(f"No AdapterFusion with name '{name}' available.")
                return

        if not exists(save_directory):
            mkdir(save_directory)
        else:
            assert isdir(save_directory), "Saving path should be a directory where the head can be saved."

        adapter_fusion_config = self.model.adapters_config.get_fusion(name)

        # Save the adapter fusion configuration
        config_dict = build_full_config(
            adapter_fusion_config,
            self.model.config,
            name=name,
            model_name=self.model.model_name,
            model_class=self.model.__class__.__name__,
        )
        self.weights_helper.save_weights_config(save_directory, config_dict, meta_dict=meta_dict)

        # Save head weights
        filter_func = self.filter_func(name)
        self.weights_helper.save_weights(save_directory, filter_func)

    def load_from_state_dict(self, state_dict, name, load_as=None, loading_info=None, start_prefix=""):
        """
        Loads the weights of a given AdapterFusion module from a state dict into the model.

        Args:
            state_dict (dict): The state dict from which to load the AdapterFusion weights.
            name (str): The name of the AdapterFusion to be loaded.
            load_as (str, optional):
                Load the AdapterFusion using this name. By default, the name with which the AdapterFusion was saved
                will be used.
            loading_info (dict, optional):
                A dictionary to which loading information (missing and unexpected keys) will be added.
            start_prefix (str, optional): A custom prefix to be ignored in the given state dict.
        """
        new_adapter_fusion_name = load_as or name
        if new_adapter_fusion_name not in self.model.adapters_config.fusions:
            raise ValueError(f"No AdapterFusion with name '{new_adapter_fusion_name}' available.")

        # Load adapter weights
        filter_func = self.filter_func(name)
        rename_func = self.rename_func(name, new_adapter_fusion_name)
        self.weights_helper.load_weights_from_state_dict(
            state_dict,
            filter_func,
            rename_func=rename_func,
            loading_info=loading_info,
            in_base_model=True,
            start_prefix=start_prefix,
        )

    def load(self, save_directory, load_as=None, loading_info=None, **kwargs):
        """
        Loads a AdapterFusion module from the given directory.

        Args:
            save_directory (str): The directory from where to load the weights.
            load_as (str, optional): Load the weights with this name. Defaults to None.

        Returns:
            Tuple[str, str]: A tuple consisting of the local file system directory from which the weights where loaded
            and the name of the loaded weights.
        """
        if not exists(join(save_directory, ADAPTERFUSION_WEIGHTS_NAME)) and not exists(
            join(save_directory, SAFE_ADAPTERFUSION_WEIGHTS_NAME)
        ):
            if self.error_on_missing:
                raise ValueError("Loading path should be a directory where AdapterFusion is saved.")
            else:
                logger.debug("No matching adapter fusion found in '{}'".format(save_directory))
                return None, None

        config = self.weights_helper.load_weights_config(save_directory)

        adapter_fusion_name = load_as or config["name"]
        if adapter_fusion_name not in self.model.adapters_config.fusions:
            self.model.add_adapter_fusion(
                adapter_fusion_name, config["config"], overwrite_ok=True, set_active=kwargs.pop("set_active", True)
            )
        else:
            logger.warning("Overwriting existing adapter fusion module '{}'".format(adapter_fusion_name))

        # Load AdapterFusion weights
        filter_func = self.filter_func(adapter_fusion_name)
        if load_as:
            rename_func = self.rename_func(config["name"], load_as)
        else:
            rename_func = None
        self.weights_helper.load_weights(
            save_directory, filter_func, rename_func=rename_func, loading_info=loading_info
        )

        return save_directory, adapter_fusion_name


class PredictionHeadLoader(WeightsLoader):
    """
    A class providing methods for saving and loading prediction head modules from the file system.

    Model classes supporting configurable head modules via config files should provide a prediction head dict at
    `model.heads` and a method `add_prediction_head(head_name, config)`.
    """

    def __init__(self, model, error_on_missing=True, convert_to_flex_head=False, use_safetensors: bool = False):
        super().__init__(
            model,
            HEAD_WEIGHTS_NAME,
            HEAD_CONFIG_NAME,
            use_safetensors=use_safetensors,
            safe_weights_name=SAFE_HEAD_WEIGHTS_NAME,
        )
        self.error_on_missing = error_on_missing
        self.convert_to_flex_head = convert_to_flex_head

    def filter_func(self, head_name):
        # ToDo remove this workaround
        if self.model.config.model_type in ["t5", "mt5"]:
            if head_name:
                return (
                    lambda x: not x.startswith("encoder")
                    and not x.startswith("decoder")
                    and not x.startswith("shared")
                    and "heads.{}".format(head_name) in x
                )
            else:
                return (
                    lambda x: not x.startswith("encoder")
                    and not x.startswith("decoder")
                    and not x.startswith("shared")
                )

        if head_name:
            return lambda x: not x.startswith(self.model.base_model_prefix) and "heads.{}".format(head_name) in x
        else:
            return lambda x: not x.startswith(self.model.base_model_prefix)

    def rename_func(self, old_name, new_name):
        return lambda k: k.replace("heads.{}".format(old_name), "heads.{}".format(new_name))

    def save(self, save_directory: str, name: str = None):
        """
        Saves a prediction head module into the given directory.

        Args:
            save_directory (str): The directory to save the weights in.
            name (str, optional): The prediction head name.
        """

        if name:
            if hasattr(self.model, "heads"):
                if name not in self.model.heads:
                    if self.error_on_missing:
                        raise ValueError(f"Unknown head_name '{name}'.")
                    else:
                        logger.debug(f"No prediction head with name '{name}' available.")
                        return
            else:
                # we haven't found a prediction head configuration, so we assume there is only one (unnamed) head
                # (e.g. this is the case if we use a 'classic' Hf model with head)
                # -> ignore the name and go on
                name = None
        if not exists(save_directory):
            mkdir(save_directory)
        else:
            assert isdir(save_directory), "Saving path should be a directory where the head can be saved."

        # if we use a custom head, save it
        if name and hasattr(self.model, "heads"):
            head = self.model.heads[name]
            head_config = head.config
        else:
            head_config = None

        # Save the adapter configuration
        config_dict = build_full_config(
            head_config,
            self.model.config,
            name=name,
            model_name=self.model.model_name,
            model_class=self.model.__class__.__name__,
            save_id2label=True,
        )
        # Add number of labels to config if present
        if head_config is None and hasattr(self.model.config, "num_labels"):
            config_dict["num_labels"] = self.model.config.num_labels

        self.weights_helper.save_weights_config(save_directory, config_dict)

        # Save head weights

        filter_func = self.filter_func(name)
        self.weights_helper.save_weights(save_directory, filter_func)

    def load(self, save_directory, load_as=None, loading_info=None, **kwargs):
        """
        Loads a prediction head module from the given directory.

        Args:
            save_directory (str): The directory from where to load the weights.
            load_as (str, optional): Load the weights with this name. Defaults to None.

        Returns:
            Tuple[str, str]: A tuple consisting of the local file system directory from which the weights where loaded
            and the name of the loaded weights.
        """
        if not exists(join(save_directory, HEAD_WEIGHTS_NAME)) and not exists(
            join(save_directory, SAFE_HEAD_WEIGHTS_NAME)
        ):
            if self.error_on_missing:
                raise ValueError("Loading path should be a directory where the head is saved.")
            else:
                logger.info("No matching prediction head found in '{}'".format(save_directory))
                return None, None
        # label2id map used to override default
        custom_id2label = kwargs.pop("id2label", None)
        if custom_id2label:
            custom_label2id = {label: id_ for id_, label in custom_id2label.items()}
        else:
            custom_label2id = None

        head_name = None
        conversion_rename_func = None

        # Load head config if available - otherwise just blindly try to load the weights
        if isfile(join(save_directory, HEAD_CONFIG_NAME)):
            config = self.weights_helper.load_weights_config(save_directory)
            # make sure that the model class of the loaded head matches the current class
            if not self.convert_to_flex_head and self.model.__class__.__name__ != config["model_class"]:
                error_msg = (
                    f"Model class '{config['model_class']}' of found prediction head does not match current model"
                    " class."
                )
                if self.error_on_missing:
                    raise ValueError(error_msg)
                else:
                    logger.warning(error_msg)
                    return None, None

            # model with flex heads
            if hasattr(self.model, "heads"):
                # load head of same model class, no conversion needed
                if self.model.__class__.__name__ == config["model_class"]:
                    head_name = load_as or config["name"]
                    head_config = config["config"]
                elif config["model_class"].endswith("ModelWithHeads"):
                    this_class = self.model.__class__.__name__.replace("AdapterModel", "")
                    other_class = config["model_class"].replace("ModelWithHeads", "")
                    if this_class == other_class:
                        head_name = load_as or config["name"]
                        head_config = config["config"]
                    else:
                        raise ValueError(
                            f"Cannot automatically convert prediction head of model class {config['model_class']} to"
                            " flex head."
                        )
                # try to convert a static head to a flex head
                elif self.convert_to_flex_head and config["model_class"] in STATIC_TO_FLEX_HEAD_MAP:
                    head_name = kwargs.pop("main_load_name", load_as)
                    if head_name is None:
                        raise ValueError(
                            "Could not identify a name for the prediction head to be loaded. Please specify 'load_as'."
                        )
                    head_config, conversion_rename_func = get_head_config_and_rename_list(
                        config["model_class"],
                        head_name,
                        custom_label2id or config.get("label2id"),
                        num_labels=config.get("num_labels"),
                    )
                else:
                    raise ValueError(
                        f"Cannot automatically convert prediction head of model class {config['model_class']} to flex"
                        " head."
                    )
                if head_name in self.model.heads:
                    logger.warning("Overwriting existing head '{}'".format(head_name))

                # make sure the label2id map is correct
                custom_label2id = custom_label2id or head_config.get("label2id", None)
                if custom_label2id:
                    head_config["id2label"] = {int(id_): label for label, id_ in custom_label2id.items()}
                    head_config["label2id"] = {label: int(id_) for label, id_ in custom_label2id.items()}

                self.model.add_prediction_head_from_config(
                    head_name, head_config, overwrite_ok=True, set_active=kwargs.pop("set_active", True)
                )
            # model with static head
            else:
                if self.convert_to_flex_head:
                    raise ValueError("Cannot set convert_flex_head on model class with static head.")
                custom_label2id = custom_label2id or config.get("label2id", None)
                if custom_label2id:
                    self.model.config.id2label = {int(id_): label for label, id_ in custom_label2id.items()}
                    self.model.config.label2id = {label: int(id_) for label, id_ in custom_label2id.items()}

        # Load head weights
        filter_func = self.filter_func(head_name)
        rename_funcs = []
        if load_as:
            rename_funcs.append(self.rename_func(config["name"], load_as))
        if conversion_rename_func:
            rename_funcs.append(conversion_rename_func)
        self.weights_helper.load_weights(
            save_directory, filter_func, rename_func=rename_funcs, loading_info=loading_info
        )

        return save_directory, head_name

    def convert_static_to_flex_head(self, state_dict, load_as="default"):
        """
        Loads a prediction head module from the given state dict, which contains a static head checkpoint.

        Args:
            state_dict (dict): The static head checkpoint from which to load the head module. Can be None.
            load_as (str, optional): Load the weights with this name. Defaults to None.

        Returns:
            Tuple[dict, dict]: A tuple consisting of the head config and the state dict of the loaded weights.
        """
        assert self.convert_to_flex_head, "load_from_state_dict() can only be used with convert_to_flex_head=True."
        assert hasattr(self.model, "heads"), "load_from_state_dict() can only be used with flex heads model class."

        if state_dict is None:
            return None, None

        conversion_rename_func = None

        original_model_class = self.model.config.architectures[0] if self.model.config.architectures else None
        if original_model_class in STATIC_TO_FLEX_HEAD_MAP:
            head_config, conversion_rename_func = get_head_config_and_rename_list(
                original_model_class,
                load_as,
                getattr(self.model.config, "label2id"),
            )
        elif self.error_on_missing:
            raise ValueError(
                f"Cannot automatically convert prediction head of model class {original_model_class} to flex head."
            )
        else:
            return None, None

        # Load head weights
        if state_dict is not None:
            new_state_dict = {}
            for k, v in state_dict.items():
                new_k = conversion_rename_func(k)
                new_state_dict[new_k] = v
        else:
            new_state_dict = None
        return head_config, new_state_dict