Fix training T5 adapter models with Trainer (#599)

src/adapters/models/__init__.py

@@ -18,7 +18,12 @@
 from .gpt2.mixin_gpt2 import GPT2ModelAdapterMixin
 from .gptj.mixin_gptj import GPTJMLPAdaptersMixin, GPTJModelAdapterMixin
 from .llama.mixin_llama import LlamaModelAdapterMixin
-from .t5.mixin_t5 import T5BlockAdaptersMixin, T5ModelAdaptersMixin, T5ModelAdaptersWithHeadsMixin
+from .t5.mixin_t5 import (
+    T5BlockAdaptersMixin,
+    T5ForCondiditionalGenerationWithHeadsMixin,
+    T5ForQuestionAnsweringWithHeadsMixin,
+    T5ModelAdaptersMixin,
+)
 from .vit.mixin_vit import ViTIntermediateAdaptersMixin, ViTModelAdaptersMixin
 from .xmod.mixin_xmod import XmodModelAdaptersMixin
 
@@ -57,8 +62,8 @@
     "RobertaModel": BertModelAdaptersMixin,
     "T5Block": T5BlockAdaptersMixin,
     "T5Model": T5ModelAdaptersMixin,
-    "T5ForConditionalGeneration": T5ModelAdaptersWithHeadsMixin,
-    "T5ForQuestionAnswering": T5ModelAdaptersWithHeadsMixin,
+    "T5ForConditionalGeneration": T5ForCondiditionalGenerationWithHeadsMixin,
+    "T5ForQuestionAnswering": T5ForQuestionAnsweringWithHeadsMixin,
     "T5EncoderModel": T5ModelAdaptersMixin,
     "ViTIntermediate": ViTIntermediateAdaptersMixin,
     "ViTModel": ViTModelAdaptersMixin,

src/adapters/models/t5/mixin_t5.py

@@ -1,5 +1,6 @@
-from typing import Iterable, Tuple
+from typing import Iterable, Optional, Tuple
 
+import torch
 import torch.nn as nn
 
 from ...methods.bottleneck import BottleneckLayer
@@ -99,5 +100,28 @@ def iter_layers(self) -> Iterable[Tuple[int, nn.Module]]:
                 yield i, layer
 
 
-class T5ModelAdaptersWithHeadsMixin(ModelWithHeadsAdaptersMixin, T5ModelAdaptersMixin):
-    pass
+# Stating "labels" and "input_ids" explicitly is required for training using Trainer class
+class T5ForCondiditionalGenerationWithHeadsMixin(ModelWithHeadsAdaptersMixin, T5ModelAdaptersMixin):
+    def forward(
+        self,
+        *args,
+        input_ids: Optional[torch.LongTensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        **kwargs,
+    ):
+        return super().forward(*args, input_ids=input_ids, labels=labels, **kwargs)
+
+
+# Stating "start_positions"/"end_positions" and "input_ids" explicitly is required for training using Trainer class
+class T5ForQuestionAnsweringWithHeadsMixin(ModelWithHeadsAdaptersMixin, T5ModelAdaptersMixin):
+    def forward(
+        self,
+        *args,
+        input_ids: Optional[torch.LongTensor] = None,
+        start_positions: Optional[torch.LongTensor] = None,
+        end_positions: Optional[torch.LongTensor] = None,
+        **kwargs,
+    ):
+        return super().forward(
+            *args, input_ids=input_ids, start_positions=start_positions, end_positions=end_positions, **kwargs
+        )