litagin commited on
Commit
a2a4aef
โ€ข
1 Parent(s): 2916d61

Add bert bi

Browse files
.gitignore CHANGED
@@ -6,7 +6,7 @@ venv/
6
 
7
  /*.yml
8
  !/default_config.yml
9
- /bert/*/*.bin
10
  /bert/*/*.h5
11
  /bert/*/*.model
12
  /bert/*/*.safetensors
 
6
 
7
  /*.yml
8
  !/default_config.yml
9
+ # /bert/*/*.bin
10
  /bert/*/*.h5
11
  /bert/*/*.model
12
  /bert/*/*.safetensors
bert/chinese-roberta-wwm-ext-large/.gitattributes DELETED
@@ -1,9 +0,0 @@
1
- *.bin.* filter=lfs diff=lfs merge=lfs -text
2
- *.lfs.* filter=lfs diff=lfs merge=lfs -text
3
- *.bin filter=lfs diff=lfs merge=lfs -text
4
- *.h5 filter=lfs diff=lfs merge=lfs -text
5
- *.tflite filter=lfs diff=lfs merge=lfs -text
6
- *.tar.gz filter=lfs diff=lfs merge=lfs -text
7
- *.ot filter=lfs diff=lfs merge=lfs -text
8
- *.onnx filter=lfs diff=lfs merge=lfs -text
9
- *.msgpack filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
 
 
 
bert/chinese-roberta-wwm-ext-large/README.md DELETED
@@ -1,57 +0,0 @@
1
- ---
2
- language:
3
- - zh
4
- tags:
5
- - bert
6
- license: "apache-2.0"
7
- ---
8
-
9
- # Please use 'Bert' related functions to load this model!
10
-
11
- ## Chinese BERT with Whole Word Masking
12
- For further accelerating Chinese natural language processing, we provide **Chinese pre-trained BERT with Whole Word Masking**.
13
-
14
- **[Pre-Training with Whole Word Masking for Chinese BERT](https://arxiv.org/abs/1906.08101)**
15
- Yiming Cui, Wanxiang Che, Ting Liu, Bing Qin, Ziqing Yang, Shijin Wang, Guoping Hu
16
-
17
- This repository is developed based on๏ผšhttps://github.com/google-research/bert
18
-
19
- You may also interested in,
20
- - Chinese BERT series: https://github.com/ymcui/Chinese-BERT-wwm
21
- - Chinese MacBERT: https://github.com/ymcui/MacBERT
22
- - Chinese ELECTRA: https://github.com/ymcui/Chinese-ELECTRA
23
- - Chinese XLNet: https://github.com/ymcui/Chinese-XLNet
24
- - Knowledge Distillation Toolkit - TextBrewer: https://github.com/airaria/TextBrewer
25
-
26
- More resources by HFL: https://github.com/ymcui/HFL-Anthology
27
-
28
- ## Citation
29
- If you find the technical report or resource is useful, please cite the following technical report in your paper.
30
- - Primary: https://arxiv.org/abs/2004.13922
31
- ```
32
- @inproceedings{cui-etal-2020-revisiting,
33
- title = "Revisiting Pre-Trained Models for {C}hinese Natural Language Processing",
34
- author = "Cui, Yiming and
35
- Che, Wanxiang and
36
- Liu, Ting and
37
- Qin, Bing and
38
- Wang, Shijin and
39
- Hu, Guoping",
40
- booktitle = "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: Findings",
41
- month = nov,
42
- year = "2020",
43
- address = "Online",
44
- publisher = "Association for Computational Linguistics",
45
- url = "https://www.aclweb.org/anthology/2020.findings-emnlp.58",
46
- pages = "657--668",
47
- }
48
- ```
49
- - Secondary: https://arxiv.org/abs/1906.08101
50
- ```
51
- @article{chinese-bert-wwm,
52
- title={Pre-Training with Whole Word Masking for Chinese BERT},
53
- author={Cui, Yiming and Che, Wanxiang and Liu, Ting and Qin, Bing and Yang, Ziqing and Wang, Shijin and Hu, Guoping},
54
- journal={arXiv preprint arXiv:1906.08101},
55
- year={2019}
56
- }
57
- ```
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
bert/chinese-roberta-wwm-ext-large/added_tokens.json DELETED
@@ -1 +0,0 @@
1
- {}
 
 
bert/chinese-roberta-wwm-ext-large/config.json DELETED
@@ -1,28 +0,0 @@
1
- {
2
- "architectures": [
3
- "BertForMaskedLM"
4
- ],
5
- "attention_probs_dropout_prob": 0.1,
6
- "bos_token_id": 0,
7
- "directionality": "bidi",
8
- "eos_token_id": 2,
9
- "hidden_act": "gelu",
10
- "hidden_dropout_prob": 0.1,
11
- "hidden_size": 1024,
12
- "initializer_range": 0.02,
13
- "intermediate_size": 4096,
14
- "layer_norm_eps": 1e-12,
15
- "max_position_embeddings": 512,
16
- "model_type": "bert",
17
- "num_attention_heads": 16,
18
- "num_hidden_layers": 24,
19
- "output_past": true,
20
- "pad_token_id": 0,
21
- "pooler_fc_size": 768,
22
- "pooler_num_attention_heads": 12,
23
- "pooler_num_fc_layers": 3,
24
- "pooler_size_per_head": 128,
25
- "pooler_type": "first_token_transform",
26
- "type_vocab_size": 2,
27
- "vocab_size": 21128
28
- }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
bert/chinese-roberta-wwm-ext-large/special_tokens_map.json DELETED
@@ -1 +0,0 @@
1
- {"unk_token": "[UNK]", "sep_token": "[SEP]", "pad_token": "[PAD]", "cls_token": "[CLS]", "mask_token": "[MASK]"}
 
 
bert/chinese-roberta-wwm-ext-large/tokenizer.json DELETED
The diff for this file is too large to render. See raw diff
 
bert/chinese-roberta-wwm-ext-large/tokenizer_config.json DELETED
@@ -1 +0,0 @@
1
- {"init_inputs": []}
 
 
bert/chinese-roberta-wwm-ext-large/vocab.txt DELETED
The diff for this file is too large to render. See raw diff
 
bert/deberta-v2-large-japanese-char-wwm/pytorch_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bf0dab8ad87bd7c22e85ec71e04f2240804fda6d33196157d6b5923af6ea1201
3
+ size 1318456639
bert/deberta-v3-large/.gitattributes DELETED
@@ -1,27 +0,0 @@
1
- *.7z filter=lfs diff=lfs merge=lfs -text
2
- *.arrow filter=lfs diff=lfs merge=lfs -text
3
- *.bin filter=lfs diff=lfs merge=lfs -text
4
- *.bin.* filter=lfs diff=lfs merge=lfs -text
5
- *.bz2 filter=lfs diff=lfs merge=lfs -text
6
- *.ftz filter=lfs diff=lfs merge=lfs -text
7
- *.gz filter=lfs diff=lfs merge=lfs -text
8
- *.h5 filter=lfs diff=lfs merge=lfs -text
9
- *.joblib filter=lfs diff=lfs merge=lfs -text
10
- *.lfs.* filter=lfs diff=lfs merge=lfs -text
11
- *.model filter=lfs diff=lfs merge=lfs -text
12
- *.msgpack filter=lfs diff=lfs merge=lfs -text
13
- *.onnx filter=lfs diff=lfs merge=lfs -text
14
- *.ot filter=lfs diff=lfs merge=lfs -text
15
- *.parquet filter=lfs diff=lfs merge=lfs -text
16
- *.pb filter=lfs diff=lfs merge=lfs -text
17
- *.pt filter=lfs diff=lfs merge=lfs -text
18
- *.pth filter=lfs diff=lfs merge=lfs -text
19
- *.rar filter=lfs diff=lfs merge=lfs -text
20
- saved_model/**/* filter=lfs diff=lfs merge=lfs -text
21
- *.tar.* filter=lfs diff=lfs merge=lfs -text
22
- *.tflite filter=lfs diff=lfs merge=lfs -text
23
- *.tgz filter=lfs diff=lfs merge=lfs -text
24
- *.xz filter=lfs diff=lfs merge=lfs -text
25
- *.zip filter=lfs diff=lfs merge=lfs -text
26
- *.zstandard filter=lfs diff=lfs merge=lfs -text
27
- *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
bert/deberta-v3-large/README.md DELETED
@@ -1,93 +0,0 @@
1
- ---
2
- language: en
3
- tags:
4
- - deberta
5
- - deberta-v3
6
- - fill-mask
7
- thumbnail: https://huggingface.co/front/thumbnails/microsoft.png
8
- license: mit
9
- ---
10
-
11
- ## DeBERTaV3: Improving DeBERTa using ELECTRA-Style Pre-Training with Gradient-Disentangled Embedding Sharing
12
-
13
- [DeBERTa](https://arxiv.org/abs/2006.03654) improves the BERT and RoBERTa models using disentangled attention and enhanced mask decoder. With those two improvements, DeBERTa out perform RoBERTa on a majority of NLU tasks with 80GB training data.
14
-
15
- In [DeBERTa V3](https://arxiv.org/abs/2111.09543), we further improved the efficiency of DeBERTa using ELECTRA-Style pre-training with Gradient Disentangled Embedding Sharing. Compared to DeBERTa, our V3 version significantly improves the model performance on downstream tasks. You can find more technique details about the new model from our [paper](https://arxiv.org/abs/2111.09543).
16
-
17
- Please check the [official repository](https://github.com/microsoft/DeBERTa) for more implementation details and updates.
18
-
19
- The DeBERTa V3 large model comes with 24 layers and a hidden size of 1024. It has 304M backbone parameters with a vocabulary containing 128K tokens which introduces 131M parameters in the Embedding layer. This model was trained using the 160GB data as DeBERTa V2.
20
-
21
-
22
- #### Fine-tuning on NLU tasks
23
-
24
- We present the dev results on SQuAD 2.0 and MNLI tasks.
25
-
26
- | Model |Vocabulary(K)|Backbone #Params(M)| SQuAD 2.0(F1/EM) | MNLI-m/mm(ACC)|
27
- |-------------------|----------|-------------------|-----------|----------|
28
- | RoBERTa-large |50 |304 | 89.4/86.5 | 90.2 |
29
- | XLNet-large |32 |- | 90.6/87.9 | 90.8 |
30
- | DeBERTa-large |50 |- | 90.7/88.0 | 91.3 |
31
- | **DeBERTa-v3-large**|128|304 | **91.5/89.0**| **91.8/91.9**|
32
-
33
-
34
- #### Fine-tuning with HF transformers
35
-
36
- ```bash
37
- #!/bin/bash
38
-
39
- cd transformers/examples/pytorch/text-classification/
40
-
41
- pip install datasets
42
- export TASK_NAME=mnli
43
-
44
- output_dir="ds_results"
45
-
46
- num_gpus=8
47
-
48
- batch_size=8
49
-
50
- python -m torch.distributed.launch --nproc_per_node=${num_gpus} \
51
- run_glue.py \
52
- --model_name_or_path microsoft/deberta-v3-large \
53
- --task_name $TASK_NAME \
54
- --do_train \
55
- --do_eval \
56
- --evaluation_strategy steps \
57
- --max_seq_length 256 \
58
- --warmup_steps 50 \
59
- --per_device_train_batch_size ${batch_size} \
60
- --learning_rate 6e-6 \
61
- --num_train_epochs 2 \
62
- --output_dir $output_dir \
63
- --overwrite_output_dir \
64
- --logging_steps 1000 \
65
- --logging_dir $output_dir
66
-
67
- ```
68
-
69
- ### Citation
70
-
71
- If you find DeBERTa useful for your work, please cite the following papers:
72
-
73
- ``` latex
74
- @misc{he2021debertav3,
75
- title={DeBERTaV3: Improving DeBERTa using ELECTRA-Style Pre-Training with Gradient-Disentangled Embedding Sharing},
76
- author={Pengcheng He and Jianfeng Gao and Weizhu Chen},
77
- year={2021},
78
- eprint={2111.09543},
79
- archivePrefix={arXiv},
80
- primaryClass={cs.CL}
81
- }
82
- ```
83
-
84
- ``` latex
85
- @inproceedings{
86
- he2021deberta,
87
- title={DEBERTA: DECODING-ENHANCED BERT WITH DISENTANGLED ATTENTION},
88
- author={Pengcheng He and Xiaodong Liu and Jianfeng Gao and Weizhu Chen},
89
- booktitle={International Conference on Learning Representations},
90
- year={2021},
91
- url={https://openreview.net/forum?id=XPZIaotutsD}
92
- }
93
- ```
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
bert/deberta-v3-large/config.json DELETED
@@ -1,22 +0,0 @@
1
- {
2
- "model_type": "deberta-v2",
3
- "attention_probs_dropout_prob": 0.1,
4
- "hidden_act": "gelu",
5
- "hidden_dropout_prob": 0.1,
6
- "hidden_size": 1024,
7
- "initializer_range": 0.02,
8
- "intermediate_size": 4096,
9
- "max_position_embeddings": 512,
10
- "relative_attention": true,
11
- "position_buckets": 256,
12
- "norm_rel_ebd": "layer_norm",
13
- "share_att_key": true,
14
- "pos_att_type": "p2c|c2p",
15
- "layer_norm_eps": 1e-7,
16
- "max_relative_positions": -1,
17
- "position_biased_input": false,
18
- "num_attention_heads": 16,
19
- "num_hidden_layers": 24,
20
- "type_vocab_size": 0,
21
- "vocab_size": 128100
22
- }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
bert/deberta-v3-large/generator_config.json DELETED
@@ -1,22 +0,0 @@
1
- {
2
- "model_type": "deberta-v2",
3
- "attention_probs_dropout_prob": 0.1,
4
- "hidden_act": "gelu",
5
- "hidden_dropout_prob": 0.1,
6
- "hidden_size": 1024,
7
- "initializer_range": 0.02,
8
- "intermediate_size": 4096,
9
- "max_position_embeddings": 512,
10
- "relative_attention": true,
11
- "position_buckets": 256,
12
- "norm_rel_ebd": "layer_norm",
13
- "share_att_key": true,
14
- "pos_att_type": "p2c|c2p",
15
- "layer_norm_eps": 1e-7,
16
- "max_relative_positions": -1,
17
- "position_biased_input": false,
18
- "num_attention_heads": 16,
19
- "num_hidden_layers": 12,
20
- "type_vocab_size": 0,
21
- "vocab_size": 128100
22
- }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
bert/deberta-v3-large/tokenizer_config.json DELETED
@@ -1,4 +0,0 @@
1
- {
2
- "do_lower_case": false,
3
- "vocab_type": "spm"
4
- }