asahi417 commited on
Commit
544284f
1 Parent(s): abc8b66

model update

Browse files
README.md ADDED
@@ -0,0 +1,150 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ datasets:
3
+ - tner/ontonotes5
4
+ metrics:
5
+ - f1
6
+ - precision
7
+ - recall
8
+ model-index:
9
+ - name: tner/deberta-v3-large-ontonotes5
10
+ results:
11
+ - task:
12
+ name: Token Classification
13
+ type: token-classification
14
+ dataset:
15
+ name: tner/ontonotes5
16
+ type: tner/ontonotes5
17
+ args: tner/ontonotes5
18
+ metrics:
19
+ - name: F1
20
+ type: f1
21
+ value: 0.9069623608411381
22
+ - name: Precision
23
+ type: precision
24
+ value: 0.902100360312857
25
+ - name: Recall
26
+ type: recall
27
+ value: 0.9118770542773386
28
+ - name: F1 (macro)
29
+ type: f1_macro
30
+ value: 0.834586960779896
31
+ - name: Precision (macro)
32
+ type: precision_macro
33
+ value: 0.8237351069457466
34
+ - name: Recall (macro)
35
+ type: recall_macro
36
+ value: 0.8475169311172334
37
+ - name: F1 (entity span)
38
+ type: f1_entity_span
39
+ value: 0.9267538434352359
40
+ - name: Precision (entity span)
41
+ type: precision_entity_span
42
+ value: 0.9217857456718517
43
+ - name: Recall (entity span)
44
+ type: recall_entity_span
45
+ value: 0.9317757839566492
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-ontonotes5
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/ontonotes5](https://huggingface.co/datasets/tner/ontonotes5) 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): 0.9069623608411381
59
+ - Precision (micro): 0.902100360312857
60
+ - Recall (micro): 0.9118770542773386
61
+ - F1 (macro): 0.834586960779896
62
+ - Precision (macro): 0.8237351069457466
63
+ - Recall (macro): 0.8475169311172334
64
+
65
+ The per-entity breakdown of the F1 score on the test set are below:
66
+ - cardinal_number: 0.853475935828877
67
+ - date: 0.8815545959284392
68
+ - event: 0.8030303030303031
69
+ - facility: 0.7896678966789669
70
+ - geopolitical_area: 0.9650033867690223
71
+ - group: 0.9337209302325581
72
+ - language: 0.8372093023255814
73
+ - law: 0.6756756756756757
74
+ - location: 0.7624020887728459
75
+ - money: 0.8818897637795275
76
+ - ordinal_number: 0.8635235732009926
77
+ - organization: 0.914952751528627
78
+ - percent: 0.9
79
+ - person: 0.9609866599546942
80
+ - product: 0.7901234567901234
81
+ - quantity: 0.8161434977578474
82
+ - time: 0.674364896073903
83
+ - work_of_art: 0.7188405797101449
84
+
85
+ For F1 scores, the confidence interval is obtained by bootstrap as below:
86
+ - F1 (micro):
87
+ - 90%: [0.9019409960743083, 0.911751130722053]
88
+ - 95%: [0.9010822890967028, 0.9125611412371442]
89
+ - F1 (macro):
90
+ - 90%: [0.9019409960743083, 0.911751130722053]
91
+ - 95%: [0.9010822890967028, 0.9125611412371442]
92
+
93
+ Full evaluation can be found at [metric file of NER](https://huggingface.co/tner/deberta-v3-large-ontonotes5/raw/main/eval/metric.json)
94
+ and [metric file of entity span](https://huggingface.co/tner/deberta-v3-large-ontonotes5/raw/main/eval/metric_span.json).
95
+
96
+ ### Usage
97
+ This model can be used through the [tner library](https://github.com/asahi417/tner). Install the library via pip
98
+ ```shell
99
+ pip install tner
100
+ ```
101
+ and activate model as below.
102
+ ```python
103
+ from tner import TransformersNER
104
+ model = TransformersNER("tner/deberta-v3-large-ontonotes5")
105
+ model.predict(["Jacob Collier is a Grammy awarded English artist from London"])
106
+ ```
107
+ It can be used via transformers library but it is not recommended as CRF layer is not supported at the moment.
108
+
109
+ ### Training hyperparameters
110
+
111
+ The following hyperparameters were used during training:
112
+ - dataset: ['tner/ontonotes5']
113
+ - dataset_split: train
114
+ - dataset_name: None
115
+ - local_dataset: None
116
+ - model: microsoft/deberta-v3-large
117
+ - crf: True
118
+ - max_length: 128
119
+ - epoch: 15
120
+ - batch_size: 16
121
+ - lr: 1e-05
122
+ - random_seed: 42
123
+ - gradient_accumulation_steps: 4
124
+ - weight_decay: 1e-07
125
+ - lr_warmup_step_ratio: 0.1
126
+ - max_grad_norm: 10.0
127
+
128
+ The full configuration can be found at [fine-tuning parameter file](https://huggingface.co/tner/deberta-v3-large-ontonotes5/raw/main/trainer_config.json).
129
+
130
+ ### Reference
131
+ If you use any resource from T-NER, please consider to cite our [paper](https://aclanthology.org/2021.eacl-demos.7/).
132
+
133
+ ```
134
+
135
+ @inproceedings{ushio-camacho-collados-2021-ner,
136
+ title = "{T}-{NER}: An All-Round Python Library for Transformer-based Named Entity Recognition",
137
+ author = "Ushio, Asahi and
138
+ Camacho-Collados, Jose",
139
+ booktitle = "Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: System Demonstrations",
140
+ month = apr,
141
+ year = "2021",
142
+ address = "Online",
143
+ publisher = "Association for Computational Linguistics",
144
+ url = "https://aclanthology.org/2021.eacl-demos.7",
145
+ doi = "10.18653/v1/2021.eacl-demos.7",
146
+ pages = "53--62",
147
+ 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.",
148
+ }
149
+
150
+ ```
config.json CHANGED
@@ -1,5 +1,5 @@
1
  {
2
- "_name_or_path": "tner_ckpt/ontonotes5_deberta_v3_large/best_model",
3
  "architectures": [
4
  "DebertaV2ForTokenClassification"
5
  ],
 
1
  {
2
+ "_name_or_path": "tner_ckpt/ontonotes5_deberta_v3_large/model_ayzfwn/epoch_5",
3
  "architectures": [
4
  "DebertaV2ForTokenClassification"
5
  ],
eval/metric.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"micro/f1": 0.9069623608411381, "micro/f1_ci": {"90": [0.9019409960743083, 0.911751130722053], "95": [0.9010822890967028, 0.9125611412371442]}, "micro/recall": 0.9118770542773386, "micro/precision": 0.902100360312857, "macro/f1": 0.834586960779896, "macro/f1_ci": {"90": [0.8209804467128787, 0.8467249999163714], "95": [0.8183331796999228, 0.8488775156564468]}, "macro/recall": 0.8475169311172334, "macro/precision": 0.8237351069457466, "per_entity_metric": {"cardinal_number": {"f1": 0.853475935828877, "f1_ci": {"90": [0.8330497989081731, 0.8718031833104767], "95": [0.8292140789883788, 0.875]}, "precision": 0.853475935828877, "recall": 0.853475935828877}, "date": {"f1": 0.8815545959284392, "f1_ci": {"90": [0.8680905896697763, 0.8941108258988675], "95": [0.8660710010497302, 0.8963422156536515]}, "precision": 0.8713414634146341, "recall": 0.8920099875156055}, "event": {"f1": 0.8030303030303031, "f1_ci": {"90": [0.7258064516129032, 0.8676990493166965], "95": [0.7121147919876734, 0.8799999999999999]}, "precision": 0.7681159420289855, "recall": 0.8412698412698413}, "facility": {"f1": 0.7896678966789669, "f1_ci": {"90": [0.733052273663965, 0.8432094752717251], "95": [0.725635331440975, 0.8512490002665958]}, "precision": 0.7867647058823529, "recall": 0.7925925925925926}, "geopolitical_area": {"f1": 0.9650033867690223, "f1_ci": {"90": [0.959691207666892, 0.9698219691594531], "95": [0.9588066921705949, 0.970889877651981]}, "precision": 0.9762448606669712, "recall": 0.9540178571428571}, "group": {"f1": 0.9337209302325581, "f1_ci": {"90": [0.9186619111684506, 0.9476157489809728], "95": [0.9160821541280667, 0.9494624796759169]}, "precision": 0.9135381114903299, "recall": 0.9548156956004756}, "language": {"f1": 0.8372093023255814, "f1_ci": {"90": [0.7317073170731707, 0.9230769230769231], "95": [0.7058823529411765, 0.9411764705882353]}, "precision": 0.8571428571428571, "recall": 0.8181818181818182}, "law": {"f1": 0.6756756756756757, "f1_ci": {"90": [0.5507246376811594, 0.7858225108225106], "95": [0.5194564694564695, 0.8048859676908459]}, "precision": 0.7352941176470589, "recall": 0.625}, "location": {"f1": 0.7624020887728459, "f1_ci": {"90": [0.7159090909090909, 0.8052948402948403], "95": [0.7052152460836314, 0.8146637915859885]}, "precision": 0.7156862745098039, "recall": 0.8156424581005587}, "money": {"f1": 0.8818897637795275, "f1_ci": {"90": [0.8486692716920033, 0.9141318240683817], "95": [0.8442179400941717, 0.9222977161045344]}, "precision": 0.8722741433021807, "recall": 0.89171974522293}, "ordinal_number": {"f1": 0.8635235732009926, "f1_ci": {"90": [0.8323637555580891, 0.8932087017273362], "95": [0.824452588940783, 0.8984027892294282]}, "precision": 0.8365384615384616, "recall": 0.8923076923076924}, "organization": {"f1": 0.914952751528627, "f1_ci": {"90": [0.9040703443911496, 0.9256022910341289], "95": [0.9023951322353315, 0.9270120526085166]}, "precision": 0.9129229062673322, "recall": 0.916991643454039}, "percent": {"f1": 0.9, "f1_ci": {"90": [0.8643685509834927, 0.9322760755538995], "95": [0.858823101292919, 0.9376177922077923]}, "precision": 0.8974358974358975, "recall": 0.9025787965616046}, "person": {"f1": 0.9609866599546942, "f1_ci": {"90": [0.9549051480350558, 0.9671870030430941], "95": [0.9536727221285938, 0.9684061859032064]}, "precision": 0.961712846347607, "recall": 0.960261569416499}, "product": {"f1": 0.7901234567901234, "f1_ci": {"90": [0.7134465675866949, 0.8588271382172501], "95": [0.6945750544365813, 0.8689675697865352]}, "precision": 0.7441860465116279, "recall": 0.8421052631578947}, "quantity": {"f1": 0.8161434977578474, "f1_ci": {"90": [0.7536101904204847, 0.8716280602159256], "95": [0.7377543859649123, 0.8820588235294119]}, "precision": 0.7711864406779662, "recall": 0.8666666666666667}, "time": {"f1": 0.674364896073903, "f1_ci": {"90": [0.6221192233423761, 0.721523216148403], "95": [0.6117517618294261, 0.7317103734947124]}, "precision": 0.6606334841628959, "recall": 0.6886792452830188}, "work_of_art": {"f1": 0.7188405797101449, "f1_ci": {"90": [0.6665750915750914, 0.7664173228346457], "95": [0.6527626811594202, 0.7781568101472706]}, "precision": 0.6927374301675978, "recall": 0.7469879518072289}}}
eval/metric_span.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"micro/f1": 0.9267538434352359, "micro/f1_ci": {"90": [0.922241598332788, 0.9310239217571861], "95": [0.9214101526505009, 0.9317962848686707]}, "micro/recall": 0.9317757839566492, "micro/precision": 0.9217857456718517, "macro/f1": 0.9267538434352359, "macro/f1_ci": {"90": [0.922241598332788, 0.9310239217571861], "95": [0.9214101526505009, 0.9317962848686707]}, "macro/recall": 0.9317757839566492, "macro/precision": 0.9217857456718517}
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 sha256:6623273842f111141efed7506f70815efe36cdb90960ed934f41edbb979513fd
3
- size 1736332335
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bf457f85553cfc6301d4ec312cc1723387adcca7d8d8e001b5f52c357b4bac45
3
+ size 1736337839
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/ontonotes5_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": "tner_ckpt/ontonotes5_deberta_v3_large/model_ayzfwn/epoch_5",
8
  "pad_token": "[PAD]",
9
  "sep_token": "[SEP]",
10
  "sp_model_kwargs": {},
trainer_config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"dataset": ["tner/ontonotes5"], "dataset_split": "train", "dataset_name": null, "local_dataset": null, "model": "microsoft/deberta-v3-large", "crf": true, "max_length": 128, "epoch": 15, "batch_size": 16, "lr": 1e-05, "random_seed": 42, "gradient_accumulation_steps": 4, "weight_decay": 1e-07, "lr_warmup_step_ratio": 0.1, "max_grad_norm": 10.0}