Upload 3 files

SciDeBERTa-full is trained with S2ORC dataset which includes all categories and not only abstract but also body texts, the total size of the dataset is about 260GB.
SciDeBERTa-full is trained from scratch and the total number of training steps is 500K.
The experimental result with SciDeBERTa-full model is as below picture.



config.json

@@ -0,0 +1,25 @@
+{
+  "model_type": "deberta-v2",
+  "attention_head_size": 64,
+  "attention_probs_dropout_prob": 0.1,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 768,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "layer_norm_eps": 1e-07,
+  "max_position_embeddings": 512,
+  "max_relative_positions": -1,
+  "norm_rel_ebd": "layer_norm",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "padding_idx": 0,
+  "pos_att_type": "p2c|c2p",
+  "position_biased_input": false,
+  "position_buckets": 256,
+  "relative_attention": true,
+  "share_att_key": true,
+  "type_vocab_size": 0,
+  "vocab_size": 128100
+}
+

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:614b652d8e1518ca71f2b5792b169357f231e0b6818e756542a4b9634ebd7295
+size 369920519

spm.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c679fbf93643d19aab7ee10c0b99e460bdbc02fedf34b92b05af343b4af586fd
+size 2464616