File size: 2,666 Bytes
f42551d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
base_model: SALT-NLP/FLANG-ELECTRA
tags:
- generated_from_trainer
metrics:
- accuracy
- f1
- precision
- recall
model-index:
- name: FLANG-ELECTRA_flang-bert
  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. -->

# FLANG-ELECTRA_flang-bert

This model is a fine-tuned version of [SALT-NLP/FLANG-ELECTRA](https://huggingface.co/SALT-NLP/FLANG-ELECTRA) on an unknown dataset.
It achieves the following results on the evaluation set:
- Loss: 0.5930
- Accuracy: 0.8705
- F1: 0.8717
- Precision: 0.8772
- Recall: 0.8705

## 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: 0.0001
- train_batch_size: 32
- eval_batch_size: 32
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 1000
- num_epochs: 25

### Training results

| Training Loss | Epoch | Step | Validation Loss | Accuracy | F1     | Precision | Recall |
|:-------------:|:-----:|:----:|:---------------:|:--------:|:------:|:---------:|:------:|
| 0.6377        | 1.0   | 181  | 0.5174          | 0.8003   | 0.7860 | 0.8080    | 0.8003 |
| 0.4035        | 2.0   | 362  | 0.4221          | 0.8580   | 0.8578 | 0.8611    | 0.8580 |
| 0.2395        | 3.0   | 543  | 0.4535          | 0.8580   | 0.8560 | 0.8592    | 0.8580 |
| 0.231         | 4.0   | 724  | 0.4335          | 0.8658   | 0.8657 | 0.8659    | 0.8658 |
| 0.3369        | 5.0   | 905  | 0.5608          | 0.8081   | 0.8057 | 0.8151    | 0.8081 |
| 0.2203        | 6.0   | 1086 | 0.5002          | 0.8705   | 0.8691 | 0.8706    | 0.8705 |
| 0.239         | 7.0   | 1267 | 0.6676          | 0.8128   | 0.8125 | 0.8338    | 0.8128 |
| 0.0938        | 8.0   | 1448 | 0.5930          | 0.8705   | 0.8717 | 0.8772    | 0.8705 |
| 0.1329        | 9.0   | 1629 | 0.5017          | 0.8580   | 0.8571 | 0.8572    | 0.8580 |
| 0.3598        | 10.0  | 1810 | 0.5126          | 0.8690   | 0.8675 | 0.8698    | 0.8690 |
| 0.1615        | 11.0  | 1991 | 0.5945          | 0.8612   | 0.8605 | 0.8606    | 0.8612 |
| 0.0923        | 12.0  | 2172 | 0.8213          | 0.8268   | 0.8292 | 0.8450    | 0.8268 |
| 0.1296        | 13.0  | 2353 | 0.8647          | 0.8580   | 0.8586 | 0.8611    | 0.8580 |


### Framework versions

- Transformers 4.37.0
- Pytorch 2.1.2
- Datasets 2.1.0
- Tokenizers 0.15.1