Fix GenerationMixin warning for AdapterModel classes (#787)

Fixes the following warning for all AdapterModels:
```
LlamaAdapterModel has generative capabilities, as `prepare_inputs_for_generation` is explicitly overwritten. However, it doesn't directly inherit from `GenerationMixin`. From 👉v4.50👈 onwards, `PreTrainedModel` will NOT inherit from `GenerationMixin`, and this model will lose the ability to call `generate` and other related functions.
```

src/adapters/models/bart/adapter_model.py

@@ -1,5 +1,6 @@
 import torch
 
+from transformers.generation import GenerationMixin
 from transformers.models.bart.modeling_bart import (
     BART_INPUTS_DOCSTRING,
     BART_START_DOCSTRING,
@@ -18,7 +19,9 @@
 @add_start_docstrings(
     "BART Model with the option to add multiple flexible prediction heads on top.", BART_START_DOCSTRING
 )
-class BartAdapterModel(EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, BartPreTrainedModel):
+class BartAdapterModel(
+    EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, BartPreTrainedModel, GenerationMixin
+):
     _tied_weights_keys = [
         "encoder.embed_tokens.weight",
         "decoder.embed_tokens.weight",

src/adapters/models/bert/adapter_model.py

@@ -1,3 +1,4 @@
+from transformers.generation import GenerationMixin
 from transformers.models.bert.modeling_bert import (
     BERT_INPUTS_DOCSTRING,
     BERT_START_DOCSTRING,
@@ -16,7 +17,9 @@
     """Bert Model transformer with the option to add multiple flexible heads on top.""",
     BERT_START_DOCSTRING,
 )
-class BertAdapterModel(EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, BertPreTrainedModel):
+class BertAdapterModel(
+    EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, BertPreTrainedModel, GenerationMixin
+):
 
     head_types = [
         "classification",

src/adapters/models/bert_generation/adapter_model.py

@@ -1,3 +1,4 @@
+from transformers.generation import GenerationMixin
 from transformers.models.bert_generation.modeling_bert_generation import (
     BERT_GENERATION_INPUTS_DOCSTRING,
     BERT_GENERATION_START_DOCSTRING,
@@ -17,7 +18,7 @@
     BERT_GENERATION_START_DOCSTRING,
 )
 class BertGenerationAdapterModel(
-    EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, BertGenerationPreTrainedModel
+    EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, BertGenerationPreTrainedModel, GenerationMixin
 ):
     _keys_to_ignore_on_load_unexpected = [r"lm_head.bias"]
 

src/adapters/models/distilbert/adapter_model.py

@@ -1,5 +1,6 @@
 import torch.nn as nn
 
+from transformers.generation import GenerationMixin
 from transformers.models.distilbert.modeling_distilbert import (
     DISTILBERT_INPUTS_DOCSTRING,
     DISTILBERT_START_DOCSTRING,
@@ -18,7 +19,7 @@
     DISTILBERT_START_DOCSTRING,
 )
 class DistilBertAdapterModel(
-    EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, DistilBertPreTrainedModel
+    EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, DistilBertPreTrainedModel, GenerationMixin
 ):
     head_types = [
         "classification",

src/adapters/models/electra/adapter_model.py

@@ -1,3 +1,4 @@
+from transformers.generation import GenerationMixin
 from transformers.models.electra.modeling_electra import (
     ELECTRA_INPUTS_DOCSTRING,
     ELECTRA_START_DOCSTRING,
@@ -16,7 +17,9 @@
     """Electra Model transformer with the option to add multiple flexible heads on top.""",
     ELECTRA_START_DOCSTRING,
 )
-class ElectraAdapterModel(EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, ElectraPreTrainedModel):
+class ElectraAdapterModel(
+    EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, ElectraPreTrainedModel, GenerationMixin
+):
 
     head_types = [
         "classification",

src/adapters/models/gpt2/adapter_model.py

@@ -2,6 +2,7 @@
 
 import torch
 
+from transformers.generation import GenerationMixin
 from transformers.models.gpt2.modeling_gpt2 import GPT2_START_DOCSTRING, GPT2Model, GPT2PreTrainedModel
 from transformers.utils import add_start_docstrings
 
@@ -25,7 +26,9 @@
 """,
     GPT2_START_DOCSTRING,
 )
-class GPT2AdapterModel(EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, GPT2PreTrainedModel):
+class GPT2AdapterModel(
+    EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, GPT2PreTrainedModel, GenerationMixin
+):
     head_types = [
         "classification",
         "multilabel_classification",

src/adapters/models/gptj/adapter_model.py

@@ -2,6 +2,7 @@
 
 import torch
 
+from transformers.generation import GenerationMixin
 from transformers.models.gptj.modeling_gptj import GPTJ_START_DOCSTRING, GPTJModel, GPTJPreTrainedModel
 from transformers.utils import add_start_docstrings
 
@@ -25,7 +26,9 @@
 """,
     GPTJ_START_DOCSTRING,
 )
-class GPTJAdapterModel(EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, GPTJPreTrainedModel):
+class GPTJAdapterModel(
+    EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, GPTJPreTrainedModel, GenerationMixin
+):
     head_types = [
         "classification",
         "multilabel_classification",

src/adapters/models/llama/adapter_model.py

@@ -3,6 +3,7 @@
 
 import torch
 
+from transformers.generation import GenerationMixin
 from transformers.models.llama.modeling_llama import LLAMA_START_DOCSTRING, LlamaModel, LlamaPreTrainedModel
 from transformers.utils import add_start_docstrings
 
@@ -26,7 +27,9 @@
 """,
     LLAMA_START_DOCSTRING,
 )
-class LlamaAdapterModel(EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, LlamaPreTrainedModel):
+class LlamaAdapterModel(
+    EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, LlamaPreTrainedModel, GenerationMixin
+):
     head_types = [
         "classification",
         "multilabel_classification",

src/adapters/models/mbart/adapter_model.py

@@ -1,5 +1,6 @@
 import torch
 
+from transformers.generation import GenerationMixin
 from transformers.models.mbart.modeling_mbart import (
     MBART_INPUTS_DOCSTRING,
     MBART_START_DOCSTRING,
@@ -19,7 +20,9 @@
 @add_start_docstrings(
     "MBART Model with the option to add multiple flexible prediction heads on top.", MBART_START_DOCSTRING
 )
-class MBartAdapterModel(EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, MBartPreTrainedModel):
+class MBartAdapterModel(
+    EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, MBartPreTrainedModel, GenerationMixin
+):
     _tied_weights_keys = [
         "encoder.embed_tokens.weight",
         "decoder.embed_tokens.weight",

src/adapters/models/mistral/adapter_model.py

@@ -2,6 +2,7 @@
 
 import torch
 
+from transformers.generation import GenerationMixin
 from transformers.models.mistral.modeling_mistral import MISTRAL_START_DOCSTRING, MistralModel, MistralPreTrainedModel
 from transformers.utils import add_start_docstrings
 
@@ -25,7 +26,9 @@
 """,
     MISTRAL_START_DOCSTRING,
 )
-class MistralAdapterModel(EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, MistralPreTrainedModel):
+class MistralAdapterModel(
+    EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, MistralPreTrainedModel, GenerationMixin
+):
     head_types = [
         "classification",
         "multilabel_classification",

src/adapters/models/mt5/adapter_model.py

@@ -2,6 +2,7 @@
 
 import torch
 
+from transformers.generation import GenerationMixin
 from transformers.models.mt5.modeling_mt5 import (
     MT5_INPUTS_DOCSTRING,
     MT5_START_DOCSTRING,
@@ -22,7 +23,9 @@
 @add_start_docstrings(
     "MT5 Model with the option to add multiple flexible prediction heads on top.", MT5_START_DOCSTRING
 )
-class MT5AdapterModel(EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, MT5PreTrainedModel):
+class MT5AdapterModel(
+    EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, MT5PreTrainedModel, GenerationMixin
+):
     _tied_weights_keys = [
         "encoder.embed_tokens.weight",
         "decoder.embed_tokens.weight",

src/adapters/models/plbart/adapter_model.py

@@ -1,5 +1,6 @@
 import torch
 
+from transformers.generation import GenerationMixin
 from transformers.models.plbart.modeling_plbart import (
     PLBART_INPUTS_DOCSTRING,
     PLBART_START_DOCSTRING,
@@ -18,7 +19,9 @@
 @add_start_docstrings(
     "PLBART Model with the option to add multiple flexible prediction heads on top.", PLBART_START_DOCSTRING
 )
-class PLBartAdapterModel(EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, PLBartPreTrainedModel):
+class PLBartAdapterModel(
+    EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, PLBartPreTrainedModel, GenerationMixin
+):
     _tied_weights_keys = [
         "encoder.embed_tokens.weight",
         "decoder.embed_tokens.weight",

src/adapters/models/roberta/adapter_model.py

@@ -1,3 +1,4 @@
+from transformers.generation import GenerationMixin
 from transformers.models.roberta.modeling_roberta import (
     ROBERTA_INPUTS_DOCSTRING,
     ROBERTA_START_DOCSTRING,
@@ -16,7 +17,9 @@
     """Roberta Model transformer with the option to add multiple flexible heads on top.""",
     ROBERTA_START_DOCSTRING,
 )
-class RobertaAdapterModel(EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, RobertaPreTrainedModel):
+class RobertaAdapterModel(
+    EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, RobertaPreTrainedModel, GenerationMixin
+):
     head_types = [
         "classification",
         "multilabel_classification",

src/adapters/models/t5/adapter_model.py

@@ -2,6 +2,7 @@
 
 import torch
 
+from transformers.generation import GenerationMixin
 from transformers.models.t5.modeling_t5 import T5_INPUTS_DOCSTRING, T5_START_DOCSTRING, T5Model, T5PreTrainedModel
 from transformers.utils import add_start_docstrings, add_start_docstrings_to_model_forward
 
@@ -15,7 +16,9 @@
 
 
 @add_start_docstrings("T5 Model with the option to add multiple flexible prediction heads on top.", T5_START_DOCSTRING)
-class T5AdapterModel(EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, T5PreTrainedModel):
+class T5AdapterModel(
+    EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, T5PreTrainedModel, GenerationMixin
+):
     _tied_weights_keys = [
         "encoder.embed_tokens.weight",
         "decoder.embed_tokens.weight",

src/adapters/models/whisper/adapter_model.py

@@ -1,6 +1,7 @@
 import torch
 
 from transformers import EncoderDecoderCache, StaticCache
+from transformers.generation import GenerationMixin
 from transformers.models.whisper.modeling_whisper import (
     WHISPER_INPUTS_DOCSTRING,
     WHISPER_START_DOCSTRING,
@@ -19,7 +20,9 @@
 @add_start_docstrings(
     "WHISPER Model with the option to add multiple flexible prediction heads on top.", WHISPER_START_DOCSTRING
 )
-class WhisperAdapterModel(EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, WhisperPreTrainedModel):
+class WhisperAdapterModel(
+    EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, WhisperPreTrainedModel, GenerationMixin
+):
     _tied_weights_keys = []
     head_types = ["seq2seq_lm"]
 

src/adapters/models/xlm_roberta/adapter_model.py

@@ -1,3 +1,4 @@
+from transformers.generation import GenerationMixin
 from transformers.models.xlm_roberta.modeling_xlm_roberta import (
     XLM_ROBERTA_INPUTS_DOCSTRING,
     XLM_ROBERTA_START_DOCSTRING,
@@ -17,7 +18,7 @@
     XLM_ROBERTA_START_DOCSTRING,
 )
 class XLMRobertaAdapterModel(
-    EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, XLMRobertaPreTrainedModel
+    EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, XLMRobertaPreTrainedModel, GenerationMixin
 ):
 
     head_types = [

src/adapters/models/xmod/adapter_model.py

@@ -2,6 +2,7 @@
 
 import torch
 
+from transformers.generation import GenerationMixin
 from transformers.models.xmod.modeling_xmod import (
     XMOD_INPUTS_DOCSTRING,
     XMOD_START_DOCSTRING,
@@ -20,7 +21,9 @@
     """X-MOD Model transformer with the option to add multiple flexible heads on top.""",
     XMOD_START_DOCSTRING,
 )
-class XmodAdapterModel(EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, XmodPreTrainedModel):
+class XmodAdapterModel(
+    EmbeddingAdaptersWrapperMixin, ModelWithFlexibleHeadsAdaptersMixin, XmodPreTrainedModel, GenerationMixin
+):
 
     head_types = [
         "classification",