Fix import structure

src/adapters/__init__.py

@@ -111,6 +111,7 @@
     "models.llama": ["LlamaAdapterModel"],
     "models.mbart": ["MBartAdapterModel"],
     "models.mistral": ["MistralAdapterModel"],
+    "models.mllama": ["MllamaAdapterModel"],
     "models.mt5": ["MT5AdapterModel"],
     "models.plbart": ["PLBartAdapterModel"],
     "models.roberta": ["RobertaAdapterModel"],
@@ -222,6 +223,7 @@
     from .models.llama import LlamaAdapterModel
     from .models.mbart import MBartAdapterModel
     from .models.mistral import MistralAdapterModel
+    from .models.mllama import MllamaAdapterModel
     from .models.mt5 import MT5AdapterModel
     from .models.plbart import PLBartAdapterModel
     from .models.roberta import RobertaAdapterModel

src/adapters/models/__init__.py

@@ -42,6 +42,17 @@
 )
 from .xmod.mixin_xmod import XmodModelAdaptersMixin
 
+from .mllama.mixin_mllama import (
+    MllamaCrossAttentionDecoderLayerAdaptersMixin,
+    MllamaSelfAttentionDecoderLayerAdaptersMixin,
+    MllamaTextCrossAttentionAdaptersMixin,
+    MllamaTextModelAdaptersMixin,
+    MllamaTextSelfAttentionAdaptersMixin,
+    MllamaVisionAttentionAdaptersMixin,
+    MllamaVisionEncoderAdaptersMixin,
+    MllamaVisionEncoderLayerAdaptersMixin,
+    MllamaVisionModelAdaptersMixin,
+)
 
 # IMPORTANT: Only add classes to this mapping that are not copied into the adapters package
 MODEL_MIXIN_MAPPING = {
@@ -109,4 +120,14 @@
     "WhisperForAudioClassification": WhisperForAudioClassificationWithHeadsMixin,
     "LlamaForQuestionAnswering": LlamaForQuestionAnsweringAdapterMixin,
     "MistralModel": MistralModelAdapterMixin,
+    # Mulitmodal Llama
+    "MllamaVisionModel": MllamaVisionModelAdaptersMixin,
+    "MllamaTextModel": MllamaTextModelAdaptersMixin,
+    "MllamaVisionEncoder": MllamaVisionEncoderAdaptersMixin,
+    "MllamaVisionAttention": MllamaVisionAttentionAdaptersMixin,
+    "MllamaTextSelfAttention": MllamaTextSelfAttentionAdaptersMixin,
+    "MllamaTextCrossAttention": MllamaTextCrossAttentionAdaptersMixin,
+    "MllamaVisionEncoderLayer": MllamaVisionEncoderLayerAdaptersMixin,
+    "MllamaSelfAttentionDecoderLayer": MllamaSelfAttentionDecoderLayerAdaptersMixin,
+    "MllamaCrossAttentionDecoderLayer": MllamaCrossAttentionDecoderLayerAdaptersMixin,
 }

src/adapters/models/auto/adapter_model.py

@@ -24,6 +24,7 @@
         ("llama", "LlamaAdapterModel"),
         ("mbart", "MBartAdapterModel"),
         ("mistral", "MistralAdapterModel"),
+        ("mllama", "MllamaAdapterModel")
         ("mt5", "MT5AdapterModel"),
         ("plbart", "PLBartAdapterModel"),
         ("roberta", "RobertaAdapterModel"),

src/adapters/wrappers/configuration.py

@@ -68,6 +68,7 @@
         "attention_probs_dropout_prob": "attention_dropout",
     },
     "xlm_roberta": {},
+    # TODO: add mllama
 }
 SUBMODEL_NAMES = {"clip": ["vision_config", "text_config"], "encoder-decoder": ["encoder", "decoder"]}