Mistral Adapters

src/adapters/__init__.py → src/adapters/__init__ 2.py

@@ -16,7 +16,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-__version__ = "1.0.0.dev0"
+__version__ = "0.2.2"
 
 from typing import TYPE_CHECKING
 
@@ -111,7 +111,6 @@
     "models.mbart": ["MBartAdapterModel"],
     "models.mistral": ["MistralAdapterModel"],
     "models.mt5": ["MT5AdapterModel"],
-    "models.plbart": ["PLBartAdapterModel"],
     "models.roberta": ["RobertaAdapterModel"],
     "models.t5": ["T5AdapterModel"],
     "models.vit": ["ViTAdapterModel"],
@@ -217,10 +216,9 @@
     from .models.gpt2 import GPT2AdapterModel
     from .models.gptj import GPTJAdapterModel
     from .models.llama import LlamaAdapterModel
-    from .models.mbart import MBartAdapterModel
     from .models.mistral import MistralAdapterModel
+    from .models.mbart import MBartAdapterModel
     from .models.mt5 import MT5AdapterModel
-    from .models.plbart import PLBartAdapterModel
     from .models.roberta import RobertaAdapterModel
     from .models.t5 import T5AdapterModel
     from .models.vit import ViTAdapterModel

src/adapters/composition.py → src/adapters/composition 2.py

@@ -1,5 +1,4 @@
 import itertools
-import warnings
 from collections.abc import Sequence
 from typing import List, Optional, Set, Tuple, Union
 
@@ -92,7 +91,7 @@ def __init__(
         self,
         *average_adapters: List[Union[AdapterCompositionBlock, str]],
         weights: Optional[List[float]] = None,
-        normalize_weights: bool = True,
+        normalize_weights: bool = True
     ):
         super().__init__(*average_adapters)
         if weights is not None:
@@ -129,7 +128,6 @@ def __init__(
         "bart",
         "mbart",
         "mt5",
-        "plbart",
         "gpt2",
         "gptj",
         "t5",
@@ -155,7 +153,7 @@ def validate_composition(adapter_composition: AdapterCompositionBlock, level=0,
                     f"Models of type {model_type} don't support adapter composition using {block_type.__name__}."
                 )
         for child in adapter_composition:
-            if not type(child) in ALLOWED_NESTINGS[type(adapter_composition)]:
+            if type(child) not in ALLOWED_NESTINGS[type(adapter_composition)]:
                 raise ValueError(f"Adapter setup is invalid. Cannot nest {child} in {adapter_composition}")
             # recursively validate children
             validate_composition(child, level=level + 1)
@@ -181,11 +179,6 @@ def parse_composition(adapter_composition, level=0, model_type=None) -> AdapterC
         else:
             return adapter_composition
     elif isinstance(adapter_composition, Sequence):
-        # Functionality of adapter-transformers v1.x
-        warnings.warn(
-            "Passing list objects for adapter activation is deprecated. Please use Stack or Fuse explicitly.",
-            category=FutureWarning,
-        )
         # for backwards compatibility
         if level == 1:
             block_class = Fuse

src/adapters/configuration/adapter_fusion_config.py

@@ -36,7 +36,7 @@ def load(cls, config: Union[dict, str], **kwargs):
             dict: The resolved adapter fusion configuration dictionary.
         """
         # currently storing AdapterFusion weights on AdapterHub is not supported.
-        config_dict = resolve_adapter_config(config, local_map=ADAPTERFUSION_CONFIG_MAP)
+        config_dict = resolve_adapter_config(config, local_map=ADAPTERFUSION_CONFIG_MAP, try_loading_from_hub=False)
         # convert back to dict to allow attr overrides
         if isinstance(config_dict, AdapterFusionConfig):
             config_dict = config_dict.to_dict()

src/adapters/configuration/model_adapters_config.py

@@ -237,6 +237,5 @@ def build_full_config(adapter_config, model_config, save_id2label=False, **kwarg
         config_dict["config"] = adapter_config.to_dict()
     else:
         config_dict["config"] = adapter_config
-    # add lib name before version to distinguish from adapter-transformers
-    config_dict["version"] = "adapters." + __version__
+    config_dict["version"] = __version__
     return config_dict

src/adapters/head_utils.py → src/adapters/head_utils 2.py

@@ -369,27 +369,6 @@
         },
         "layers": ["lm_head"],
     },
-    # PLBART
-    "PLBartForSequenceClassification": {
-        "config": {
-            "head_type": "classification",
-            "layers": 2,
-            "activation_function": "tanh",
-        },
-        "layers": [
-            None,
-            "classification_head.dense",
-            None,
-            None,
-            "classification_head.out_proj",
-        ],
-    },
-    "PLBartForConditionalGeneration": {
-        "config": {
-            "head_type": "seq2seq_lm",
-        },
-        "layers": ["lm_head"],
-    },
     # MT5
     "MT5ForConditionalGeneration": {
         "config": {
@@ -673,15 +652,7 @@
         },
         "layers": [None, "qa_outputs"],
     },
-    "LlamaForTokenClassification": {
-        "config": {
-            "head_type": "tagging",
-            "layers": 1,
-            "activation_function": None,
-        },
-        "layers": [None, "score"],
-    },
-    # Mistral
+    #Mistral
     "MistralForSequenceClassification": {
         "config": {
             "head_type": "classification",
@@ -698,14 +669,6 @@
         },
         "layers": ["lm_head"],
     },
-    "MistralForTokenClassification": {
-        "config": {
-            "head_type": "tagging",
-            "layers": 1,
-            "activation_function": None,
-        },
-        "layers": [None, "score"],
-    },
     # Electra
     "ElectraForTokenClassification": {
         "config": {

src/adapters/heads/model_mixin.py → src/adapters/heads 2/model_mixin 2.py

@@ -134,8 +134,6 @@ def tie_weights(self):
                 self = getattr(self, self.base_model_prefix)
             self._tie_encoder_decoder_weights(self.encoder, self.decoder, self.base_model_prefix)
 
-        super().tie_weights()
-
     def _resize_token_embeddings(self, new_num_tokens, pad_to_multiple_of=None):
         old_embeddings = self.get_input_embeddings()
         new_embeddings = self._get_resized_embeddings(old_embeddings, new_num_tokens, pad_to_multiple_of)
@@ -527,7 +525,7 @@ def forward_head(
         attention_mask=None,
         return_dict=False,
         context=None,
-        **kwargs,
+        **kwargs
     ):
         """
         The forward pass through a prediction head configuration. There are three ways to specify the used prediction