Upload DebertaArgClassifier

config.json

@@ -0,0 +1,36 @@
+{
+  "architectures": [
+    "DebertaArgClassifier"
+  ],
+  "auto_map": {
+    "AutoConfig": "configuration_deberta_arg_classifier.DebertaConfig",
+    "AutoModelForSequenceClassification": "modeling_deberta_arg_classifier.DebertaArgClassifier"
+  },
+  "id2label": {
+    "0": "Self-direction: thought",
+    "1": "Self-direction: action",
+    "2": "Stimulation",
+    "3": "Hedonism",
+    "4": "Achievement",
+    "5": "Power: dominance",
+    "6": "Power: resources",
+    "7": "Face",
+    "8": "Security: personal",
+    "9": "Security: societal",
+    "10": "Tradition",
+    "11": "Conformity: rules",
+    "12": "Conformity: interpersonal",
+    "13": "Humility",
+    "14": "Benevolence: caring",
+    "15": "Benevolence: dependability",
+    "16": "Universalism: concern",
+    "17": "Universalism: nature",
+    "18": "Universalism: tolerance",
+    "19": "Universalism: objectivity"
+  },
+  "label2id": null,
+  "model_type": "deberta_multilabel",
+  "number_labels": 20,
+  "torch_dtype": "float32",
+  "transformers_version": "4.26.1"
+}

configuration_deberta_arg_classifier.py

@@ -0,0 +1,8 @@
+from transformers import PretrainedConfig
+
+class DebertaConfig(PretrainedConfig):
+    model_type = "deberta_multilabel"
+
+    def __init__(self, num_labels: int=20, **kwargs):
+        self.number_labels = num_labels
+        super().__init__(**kwargs)

modeling_deberta_arg_classifier.py

@@ -0,0 +1,34 @@
+from transformers import PreTrainedModel, AutoModel
+import torch.nn as nn
+import torch
+from deberta_arg_classifier.configuration_deberta_arg_classifier import DebertaConfig
+
+
+class DebertaArgClassifier(PreTrainedModel):
+
+    config_class = DebertaConfig
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.bert = AutoModel.from_pretrained("microsoft/deberta-large")
+        self.classifier = nn.Linear(self.bert.config.hidden_size, config.number_labels)
+        self.criterion = nn.BCEWithLogitsLoss()
+
+
+    def forward(self, input_ids, attention_mask, labels=None):
+        output = self.bert(input_ids, attention_mask=attention_mask)
+        output = self._cls_embeddings(output)
+        output_cls = self.classifier(output)
+        output = torch.sigmoid(output_cls)
+        if labels is not None:
+            loss = self.cirterion(output_cls, labels)
+            return {"loss": loss, "logits": output}
+        return {"logits": output}
+
+
+    def _cls_embeddings(self, output):
+        '''Returns the embeddings corresponding to the <CLS> token of each text. '''
+
+        last_hidden_state = output[0]
+        cls_embeddings = last_hidden_state[:, 0]
+        return cls_embeddings

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7fef9869bc1b134b27c54ee1d90bc396dbecd058cdac65fadbaa6e75437e0a8a
-size 1620893171
+oid sha256:f563b917be492d981f2b7f518d2b9e5db5f2dda7be1f94ee0f77ff083723a3d4
+size 1620866665