add model

ROBERTA.py

@@ -0,0 +1,20 @@
+from transformers import PreTrainedModel
+import torch.nn as nn
+
+class CustomModel(PreTrainedModel):
+    
+    def __init__(self,config, base_model, n_label):
+        super().__init__(config)
+        self.model = base_model
+        ### New layers:
+        self.linear = nn.Linear(1024, n_label)
+        self.log_softmax = nn.LogSoftmax(dim= 1)
+
+    def forward(self,X):
+        
+        bert_output = self.model(**X)
+        # sequence_output has the following shape: (batch_size, sequence_length, 1024)
+        output = self.linear(bert_output[1]) ## 
+        output = self.log_softmax(output)
+
+        return output

config.json

@@ -0,0 +1,32 @@
+{
+  "_name_or_path": "klue/roberta-large",
+  "architectures": [
+    "CustomModel"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "auto_map": {
+    "AutoModel": "ROBERTA.CustomModel"
+  },
+  "bos_token_id": 0,
+  "classifier_dropout": null,
+  "eos_token_id": 2,
+  "gradient_checkpointing": false,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 1024,
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "layer_norm_eps": 1e-05,
+  "max_position_embeddings": 514,
+  "model_type": "roberta",
+  "num_attention_heads": 16,
+  "num_hidden_layers": 24,
+  "pad_token_id": 1,
+  "position_embedding_type": "absolute",
+  "tokenizer_class": "BertTokenizer",
+  "torch_dtype": "float32",
+  "transformers_version": "4.21.3",
+  "type_vocab_size": 1,
+  "use_cache": true,
+  "vocab_size": 32000
+}

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:43d8d1cbac442c4ab0a1d68769796524cc178d0c1a2571da999d6a86b894f11d
+size 1346802925