File size: 5,605 Bytes
d9274d5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
89fe2a5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d9274d5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
---
license: mit
tags:
- generated_from_trainer
model-index:
- name: multiCorp_5e-05_0404
  results: []
---

<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->

# multiCorp_5e-05_0404

This model is a fine-tuned version of [microsoft/BiomedNLP-PubMedBERT-base-uncased-abstract-fulltext](https://huggingface.co/microsoft/BiomedNLP-PubMedBERT-base-uncased-abstract-fulltext) on the None dataset.
It achieves the following results on the evaluation set:
- eval_loss: 0.0657
- eval_precision: 0.6398
- eval_recall: 0.6267
- eval_f1: 0.6332
- eval_accuracy: 0.9847
- eval_runtime: 39.7302
- eval_samples_per_second: 32.544
- eval_steps_per_second: 2.039
- epoch: 3.41
- step: 1100


Multi Corp Training,

  model = AutoModelForTokenClassification.from_pretrained(
      "microsoft/BiomedNLP-PubMedBERT-base-uncased-abstract-fulltext", num_labels=41, id2label=id2label, label2id=label2id
  )

  training_args = TrainingArguments(
      report_to = 'wandb',                     # enable logging to W&B
      output_dir = runname,    # output directory/ name for huggingface hub
      learning_rate=5e-5,
      per_device_train_batch_size=16,
      per_device_eval_batch_size=16,
      weight_decay=0.01,
      evaluation_strategy = 'steps',          # check evaluation metrics at each epoch
      max_steps = 2000,
      logging_steps = 25,                    # we will log every 25 steps
      eval_steps = 25,                      # we will perform evaluation every 25 steps
      save_steps = 25,
      load_best_model_at_end=True,
      metric_for_best_model = 'eval_loss',
      greater_is_better = False,
      push_to_hub=True,
      run_name = runname             # name of the W&B run
  )

  trainer = Trainer(
      model=model,
      args=training_args,
      train_dataset=tokenized_data["train"],
      eval_dataset=tokenized_data["validation"],
      tokenizer=tokenizer,
      data_collator=data_collator,
      compute_metrics=compute_metrics,
      callbacks = [EarlyStoppingCallback(early_stopping_patience=6)]
  )

[1101/2000 1:00:33 < 49:32, 0.30 it/s, Epoch 3.41/7] 

25 	0.836100 	0.201612 	0.000000 	0.000000 	0.000000 	0.973546
50 	0.149500 	0.154239 	0.233246 	0.124420 	0.162277 	0.972420
75 	0.136300 	0.138105 	0.145299 	0.094708 	0.114671 	0.972385
100 	0.129900 	0.123477 	0.425243 	0.203343 	0.275126 	0.975886
125 	0.103100 	0.118570 	0.297553 	0.321727 	0.309168 	0.974136
150 	0.117300 	0.113230 	0.393373 	0.214949 	0.277995 	0.977039
175 	0.117500 	0.106183 	0.320082 	0.291551 	0.305151 	0.975930
200 	0.093800 	0.102443 	0.353604 	0.291551 	0.319593 	0.975297
225 	0.091900 	0.105976 	0.446684 	0.318942 	0.372156 	0.977127
250 	0.088700 	0.093393 	0.439173 	0.335190 	0.380200 	0.977734
275 	0.113300 	0.097715 	0.522222 	0.218199 	0.307793 	0.977637
300 	0.092900 	0.085730 	0.473552 	0.349118 	0.401924 	0.979405
325 	0.085700 	0.091731 	0.380009 	0.409471 	0.394190 	0.976960
350 	0.081700 	0.086656 	0.554161 	0.389508 	0.457470 	0.980162
375 	0.062400 	0.083441 	0.538000 	0.374652 	0.441708 	0.980769
400 	0.077500 	0.085072 	0.486742 	0.477252 	0.481950 	0.978869
425 	0.073000 	0.078521 	0.516658 	0.467967 	0.491108 	0.981103
450 	0.081000 	0.077073 	0.552381 	0.430826 	0.484090 	0.981288
475 	0.075100 	0.078478 	0.483887 	0.446147 	0.464251 	0.980408
500 	0.062800 	0.073298 	0.550633 	0.484680 	0.515556 	0.982247
525 	0.060600 	0.069571 	0.542723 	0.536676 	0.539683 	0.982608
550 	0.063900 	0.071559 	0.539832 	0.506500 	0.522635 	0.981983
575 	0.060700 	0.068333 	0.564646 	0.519034 	0.540881 	0.982546
600 	0.062900 	0.072810 	0.602013 	0.416435 	0.492316 	0.981886
625 	0.051300 	0.071469 	0.550901 	0.525070 	0.537675 	0.982335
650 	0.059500 	0.067657 	0.553466 	0.478180 	0.513076 	0.982528
675 	0.047500 	0.067443 	0.594739 	0.566852 	0.580461 	0.983663
700 	0.052100 	0.065269 	0.564447 	0.546890 	0.555529 	0.983039
725 	0.041500 	0.067790 	0.593516 	0.552461 	0.572253 	0.983672
750 	0.046300 	0.067922 	0.609038 	0.538069 	0.571358 	0.983461
775 	0.054300 	0.064636 	0.646725 	0.582173 	0.612753 	0.984499
800 	0.049500 	0.067722 	0.650905 	0.517642 	0.576674 	0.983830
825 	0.043100 	0.069327 	0.630043 	0.471216 	0.539177 	0.982880
850 	0.048000 	0.063814 	0.631025 	0.528784 	0.575398 	0.984068
875 	0.042500 	0.064527 	0.644913 	0.582637 	0.612195 	0.984543
900 	0.043500 	0.065475 	0.608295 	0.490251 	0.542931 	0.983522
925 	0.039200 	0.066043 	0.635938 	0.566852 	0.599411 	0.984323
950 	0.046800 	0.062491 	0.646930 	0.547818 	0.593263 	0.984719
975 	0.043700 	0.061204 	0.634625 	0.585422 	0.609032 	0.984543
1000 	0.032000 	0.066377 	0.643390 	0.560353 	0.599007 	0.984349
1025 	0.038100 	0.064764 	0.666482 	0.559424 	0.608279 	0.984745
1050 	0.035300 	0.065642 	0.635359 	0.587279 	0.610374 	0.984464
1075 	0.032800 	0.064835 	0.657262 	0.584030 	0.618486 	0.984587
1100 	0.031700 	0.065726 	0.639810 	0.626741 	0.633208 	0.984710

## Model description

More information needed

## Intended uses & limitations

More information needed

## Training and evaluation data

More information needed

## Training procedure

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 5e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- training_steps: 2000

### Framework versions

- Transformers 4.27.4
- Pytorch 2.0.0+cu118
- Datasets 2.11.0
- Tokenizers 0.13.2