Remove deprecated functionality (#714)

Changes in this PR:
- Remove deprecated add_fusion() & train_fusion() methods
- Remove deprecated support for passing lists to adapter activation
- Add "adapters." prefix to version identifier when saving adapters to
prevent confusion to adapter-transformers versions

Author: calpt

examples/pytorch/adapterfusion/run_fusion_glue.py

@@ -27,7 +27,7 @@
 import numpy as np
 
 import adapters
-from adapters import AdapterArguments, AdapterTrainer
+from adapters import AdapterArguments, AdapterTrainer, Fuse
 from transformers import AutoConfig, AutoModelForSequenceClassification, AutoTokenizer, EvalPrediction, GlueDataset
 from transformers import GlueDataTrainingArguments as DataTrainingArguments
 from transformers import (
@@ -162,28 +162,26 @@ def main():
     model.load_adapter("qa/boolq@ukp", config=SeqBnConfig(), with_head=False)
     model.load_adapter("sentiment/imdb@ukp", config=SeqBnConfig(), with_head=False)
 
-    adapter_setup = [
-        [
-            "sst-2",
-            "mnli",
-            "rte",
-            "mrpc",
-            "qqp",
-            "cosmosqa",
-            "csqa",
-            "hellaswag",
-            "socialiqa",
-            "winogrande",
-            "cb",
-            "sick",
-            "scitail",
-            "boolq",
-            "imdb",
-        ]
-    ]
+    adapter_setup = Fuse(
+        "sst-2",
+        "mnli",
+        "rte",
+        "mrpc",
+        "qqp",
+        "cosmosqa",
+        "csqa",
+        "hellaswag",
+        "socialiqa",
+        "winogrande",
+        "cb",
+        "sick",
+        "scitail",
+        "boolq",
+        "imdb",
+    )
 
     # Add a fusion layer and tell the model to train fusion
-    model.add_adapter_fusion(adapter_setup[0], "dynamic")
+    model.add_adapter_fusion(adapter_setup, "dynamic")
     model.train_adapter_fusion(adapter_setup)
 
     # ~~~~~ Rest is again same as in standard training setup ~~~~~

examples/pytorch/text-generation/run_generation.py

@@ -225,7 +225,7 @@ def main():
     # Setup adapters
     if args.load_adapter:
         model.load_adapter(args.load_adapter, load_as="generation")
-        model.set_active_adapters(["generation"])
+        model.set_active_adapters("generation")
 
     if args.fp16:
         model.half()

notebooks/04_Cross_Lingual_Transfer.ipynb

@@ -761,7 +761,7 @@
       },
       "outputs": [],
       "source": [
-        "model.train_adapter([\"copa\"])"
+        "model.train_adapter(\"copa\")"
       ]
     },
     {

notebooks/08_NER_Wikiann.ipynb

@@ -136,6 +136,8 @@
       "outputs": [],
       "source": [
         "from adapters import AdapterConfig\n",
+        "from adapters.composition import Stack\n",
+        "\n",
         "target_language = \"gn\" # Choose any language that a bert-base-multilingual-cased language adapter is available for\n",
         "source_language = \"en\" # We support  \"en\", \"ja\", \"zh\", and \"ar\"\n",
         "\n",
@@ -156,7 +158,7 @@
         "    leave_out=[11],\n",
         ")\n",
         "# Set the adapters to be used in every forward pass\n",
-        "model.set_active_adapters([lang_adapter_name, \"wikiann\"])"
+        "model.set_active_adapters(Stack(lang_adapter_name, \"wikiann\"))"
       ]
     },
     {