asahi417 commited on
Commit
3b99c79
1 Parent(s): 5198b98

model update

Browse files
README.md ADDED
@@ -0,0 +1,124 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ datasets:
3
+ - conll2003
4
+ metrics:
5
+ - f1
6
+ - precision
7
+ - recall
8
+ model-index:
9
+ - name: tner/deberta-v3-large-conll2003
10
+ results:
11
+ - task:
12
+ name: Token Classification
13
+ type: token-classification
14
+ dataset:
15
+ name: conll2003
16
+ type: conll2003
17
+ args: conll2003
18
+ metrics:
19
+ - name: F1
20
+ type: f1
21
+ value: 0.9222388190844389
22
+ - name: Precision
23
+ type: precision
24
+ value: 0.9154020582592011
25
+ - name: Recall
26
+ type: recall
27
+ value: 0.9291784702549575
28
+ - name: F1 (macro)
29
+ type: f1_macro
30
+ value: 0.9043961692086329
31
+ - name: Precision (macro)
32
+ type: precision_macro
33
+ value: 0.8959854326377331
34
+ - name: Recall (macro)
35
+ type: recall_macro
36
+ value: 0.9135442454672595
37
+ - name: F1 (entity span)
38
+ type: f1_entity_span
39
+ value: 0.960570322126386
40
+ - name: Precision (entity span)
41
+ type: precision_entity_span
42
+ value: 0.9550227511375569
43
+ - name: Recall (entity span)
44
+ type: recall_entity_span
45
+ value: 0.9661827195467422
46
+
47
+ pipeline_tag: token-classification
48
+ widget:
49
+ - text: "Jacob Collier is a Grammy awarded artist from England."
50
+ example_title: "NER Example 1"
51
+ ---
52
+ # tner/deberta-v3-large-conll2003
53
+
54
+ This model is a fine-tuned version of [microsoft/deberta-v3-large](https://huggingface.co/microsoft/deberta-v3-large) on the
55
+ [tner/conll2003](https://huggingface.co/datasets/tner/conll2003) dataset.
56
+ Model fine-tuning is done via [T-NER](https://github.com/asahi417/tner)'s hyper-parameter search (see the repository
57
+ for more detail). It achieves the following results on the test set:
58
+ - F1 (micro): 0.9222388190844389
59
+ - Precision (micro): 0.9154020582592011
60
+ - Recall (micro): 0.9291784702549575
61
+ - F1 (macro): 0.9043961692086329
62
+ - Precision (macro): 0.8959854326377331
63
+ - Recall (macro): 0.9135442454672595
64
+
65
+ The per-entity breakdown of the F1 score on the test set are below:
66
+ - location: 0.9407496977025392
67
+ - organization: 0.9115486335586247
68
+ - other: 0.7920110192837466
69
+ - person: 0.9732753262896209
70
+
71
+ For F1 scores, the confidence interval is obtained by bootstrap as below:
72
+ - F1 (micro):
73
+ - 90%: [0.9157944386463721, 0.9286928993636353]
74
+ - 95%: [0.9146558483630953, 0.9297919809412201]
75
+ - F1 (macro):
76
+ - 90%: [0.9157944386463721, 0.9286928993636353]
77
+ - 95%: [0.9146558483630953, 0.9297919809412201]
78
+
79
+ Full evaluation can be found at [metric file of NER](https://huggingface.co/tner/deberta-v3-large-conll2003/raw/main/eval/metric.json)
80
+ and [metric file of entity span](https://huggingface.co/tner/deberta-v3-large-conll2003/raw/main/eval/metric_span.json).
81
+
82
+
83
+ ### Training hyperparameters
84
+
85
+ The following hyperparameters were used during training:
86
+ - dataset: ['tner/conll2003']
87
+ - dataset_split: train
88
+ - dataset_name: None
89
+ - local_dataset: None
90
+ - model: microsoft/deberta-v3-large
91
+ - crf: False
92
+ - max_length: 128
93
+ - epoch: 15
94
+ - batch_size: 16
95
+ - lr: 1e-05
96
+ - random_seed: 42
97
+ - gradient_accumulation_steps: 4
98
+ - weight_decay: None
99
+ - lr_warmup_step_ratio: 0.1
100
+ - max_grad_norm: 10.0
101
+
102
+ The full configuration can be found at [fine-tuning parameter file](https://huggingface.co/tner/deberta-v3-large-conll2003/raw/main/trainer_config.json).
103
+
104
+ ### Reference
105
+ If you use any resource from T-NER, please consider to cite our [paper](https://aclanthology.org/2021.eacl-demos.7/).
106
+
107
+ ```
108
+
109
+ @inproceedings{ushio-camacho-collados-2021-ner,
110
+ title = "{T}-{NER}: An All-Round Python Library for Transformer-based Named Entity Recognition",
111
+ author = "Ushio, Asahi and
112
+ Camacho-Collados, Jose",
113
+ booktitle = "Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: System Demonstrations",
114
+ month = apr,
115
+ year = "2021",
116
+ address = "Online",
117
+ publisher = "Association for Computational Linguistics",
118
+ url = "https://aclanthology.org/2021.eacl-demos.7",
119
+ doi = "10.18653/v1/2021.eacl-demos.7",
120
+ pages = "53--62",
121
+ abstract = "Language model (LM) pretraining has led to consistent improvements in many NLP downstream tasks, including named entity recognition (NER). In this paper, we present T-NER (Transformer-based Named Entity Recognition), a Python library for NER LM finetuning. In addition to its practical utility, T-NER facilitates the study and investigation of the cross-domain and cross-lingual generalization ability of LMs finetuned on NER. Our library also provides a web app where users can get model predictions interactively for arbitrary text, which facilitates qualitative model evaluation for non-expert programmers. We show the potential of the library by compiling nine public NER datasets into a unified format and evaluating the cross-domain and cross- lingual performance across the datasets. The results from our initial experiments show that in-domain performance is generally competitive across datasets. However, cross-domain generalization is challenging even with a large pretrained LM, which has nevertheless capacity to learn domain-specific features if fine- tuned on a combined dataset. To facilitate future research, we also release all our LM checkpoints via the Hugging Face model hub.",
122
+ }
123
+
124
+ ```
config.json CHANGED
@@ -1,5 +1,5 @@
1
  {
2
- "_name_or_path": "tner_ckpt/conll2003_deberta_v3_large/best_model",
3
  "architectures": [
4
  "DebertaV2ForTokenClassification"
5
  ],
@@ -51,7 +51,7 @@
51
  "relative_attention": true,
52
  "share_att_key": true,
53
  "torch_dtype": "float32",
54
- "transformers_version": "4.20.1",
55
  "type_vocab_size": 0,
56
  "vocab_size": 128100
57
  }
1
  {
2
+ "_name_or_path": "microsoft/deberta-v3-large",
3
  "architectures": [
4
  "DebertaV2ForTokenClassification"
5
  ],
51
  "relative_attention": true,
52
  "share_att_key": true,
53
  "torch_dtype": "float32",
54
+ "transformers_version": "4.21.1",
55
  "type_vocab_size": 0,
56
  "vocab_size": 128100
57
  }
eval/metric.json ADDED
@@ -0,0 +1 @@
 
1
+ {"micro/f1": 0.9222388190844389, "micro/f1_ci": {"90": [0.9157944386463721, 0.9286928993636353], "95": [0.9146558483630953, 0.9297919809412201]}, "micro/recall": 0.9291784702549575, "micro/precision": 0.9154020582592011, "macro/f1": 0.9043961692086329, "macro/f1_ci": {"90": [0.8967634601572929, 0.9120550234663798], "95": [0.8951146833014633, 0.9135073150082113]}, "macro/recall": 0.9135442454672595, "macro/precision": 0.8959854326377331, "per_entity_metric": {"location": {"f1": 0.9407496977025392, "f1_ci": {"90": [0.9327218512678329, 0.9484554935162461], "95": [0.9306220735040278, 0.9500172057211348]}, "precision": 0.948780487804878, "recall": 0.9328537170263789}, "organization": {"f1": 0.9115486335586247, "f1_ci": {"90": [0.9022883361987417, 0.9209639524074142], "95": [0.900931069471536, 0.9230338443865153]}, "precision": 0.8903559127439724, "recall": 0.9337748344370861}, "other": {"f1": 0.7920110192837466, "f1_ci": {"90": [0.7703470703328422, 0.8155099589242811], "95": [0.7655908711257591, 0.819674907814652]}, "precision": 0.7666666666666667, "recall": 0.8190883190883191}, "person": {"f1": 0.9732753262896209, "f1_ci": {"90": [0.9666809784122652, 0.9793078490377144], "95": [0.9648667613328867, 0.9802750787465647]}, "precision": 0.9781386633354153, "recall": 0.9684601113172542}}}
eval/metric_span.json ADDED
@@ -0,0 +1 @@
 
1
+ {"micro/f1": 0.960570322126386, "micro/f1_ci": {"90": [0.9562602448386762, 0.9649430028068969], "95": [0.9549445046545261, 0.965818354349358]}, "micro/recall": 0.9661827195467422, "micro/precision": 0.9550227511375569, "macro/f1": 0.960570322126386, "macro/f1_ci": {"90": [0.9562602448386762, 0.9649430028068969], "95": [0.9549445046545261, 0.965818354349358]}, "macro/recall": 0.9661827195467422, "macro/precision": 0.9550227511375569}
eval/prediction.validation.json ADDED
The diff for this file is too large to render. See raw diff
pytorch_model.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:1da4d8a767740cf2b4a424367ecc6e51b99d346ad542f8e294729eb9dc043760
3
- size 1736217519
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:72241d27925a8a7fbdfa8f15e0257b46d094c2122ba5417741dbd38cf7ffb48c
3
+ size 1736223023
tokenizer.json CHANGED
The diff for this file is too large to render. See raw diff
tokenizer_config.json CHANGED
@@ -4,7 +4,7 @@
4
  "do_lower_case": false,
5
  "eos_token": "[SEP]",
6
  "mask_token": "[MASK]",
7
- "name_or_path": "tner_ckpt/conll2003_deberta_v3_large/best_model",
8
  "pad_token": "[PAD]",
9
  "sep_token": "[SEP]",
10
  "sp_model_kwargs": {},
4
  "do_lower_case": false,
5
  "eos_token": "[SEP]",
6
  "mask_token": "[MASK]",
7
+ "name_or_path": "microsoft/deberta-v3-large",
8
  "pad_token": "[PAD]",
9
  "sep_token": "[SEP]",
10
  "sp_model_kwargs": {},
trainer_config.json ADDED
@@ -0,0 +1 @@
 
1
+ {"dataset": ["tner/conll2003"], "dataset_split": "train", "dataset_name": null, "local_dataset": null, "model": "microsoft/deberta-v3-large", "crf": false, "max_length": 128, "epoch": 15, "batch_size": 16, "lr": 1e-05, "random_seed": 42, "gradient_accumulation_steps": 4, "weight_decay": null, "lr_warmup_step_ratio": 0.1, "max_grad_norm": 10.0}