10k vocab

README.md

@@ -14,8 +14,16 @@ the Maldivian News Classification task https://github.com/Sofwath/DhivehiDataset
 
 - Start with MuRIL (similar to mBERT) with no Thaana vocabulary
 - Based on PanLex dictionaries, attach 1,100 Dhivehi words to Malayalam or English embeddings
-- Add remaining words and word-pieces from dv-wave to vocab.txt
-- Continue BERT pretraining on a TPU
+- Add remaining words and word-pieces from BertWordPieceTokenizer / vocab.txt
+- Continue BERT pretraining
+
+## Performance
+
+- mBERT: 52%
+- dv-wave (ELECTRA, 30k vocab): 89%
+- dv-muril (10k vocab) before BERT pretraining step: 89.8%
+- previous dv-muril (30k vocab): 90.7%
+- dv-muril (10k vocab): 91.6%
 
 CoLab notebook: 
-https://colab.research.google.com/drive/1kPyd60EtOFLaHRVToNS-IKj29T9JCvMq?usp=sharing
+https://colab.research.google.com/drive/113o6vkLZRkm6OwhTHrvE0x6QPpavj0fn

config.json

@@ -2,17 +2,23 @@
   "architectures": [
     "BertForMaskedLM"
   ],
-  "embedding_size": 768,
-  "model_type": "bert",
   "attention_probs_dropout_prob": 0.1,
+  "embedding_size": 768,
+  "gradient_checkpointing": false,
   "hidden_act": "gelu",
   "hidden_dropout_prob": 0.1,
   "hidden_size": 768,
   "initializer_range": 0.02,
   "intermediate_size": 3072,
+  "layer_norm_eps": 1e-12,
   "max_position_embeddings": 512,
+  "model_type": "bert",
   "num_attention_heads": 12,
   "num_hidden_layers": 12,
+  "pad_token_id": 0,
+  "position_embedding_type": "absolute",
+  "transformers_version": "4.3.2",
   "type_vocab_size": 2,
-  "vocab_size": 225175
+  "use_cache": true,
+  "vocab_size": 206415
 }

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4a90a68a0c019e4090134596ca013e2c05ee045d58892b4494017b6fb0d3b972
-size 1036895067
+oid sha256:04350f04dc6d404800d3eaa95b895305c80d0d5443e41a14c64bb1c7583bb6f4
+size 979192515

tokenizer_config.json

@@ -1 +1 @@
-{"do_lower_case": false, "do_basic_tokenize": true, "never_split": null, "unk_token": "[UNK]", "sep_token": "[SEP]", "pad_token": "[PAD]", "cls_token": "[CLS]", "mask_token": "[MASK]", "tokenize_chinese_chars": true, "strip_accents": false, "lowercase": false, "special_tokens_map_file": "/root/.cache/huggingface/transformers/d66e90c2d485729cd831f5a30ce5724c25dcfbff7055265820652f3277c1412e.750293617705dd39d21eef1e163d80ca5b23e1a0ec507fff24a087c44632142c", "tokenizer_file": null, "name_or_path": "./dvadd"}
+{"do_lower_case": false, "do_basic_tokenize": true, "never_split": null, "unk_token": "[UNK]", "sep_token": "[SEP]", "pad_token": "[PAD]", "cls_token": "[CLS]", "mask_token": "[MASK]", "tokenize_chinese_chars": true, "strip_accents": false, "lowercase": false, "tokenizer_file": null}