File size: 6,722 Bytes
f969973
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
license: mit
base_model: Harveenchadha/vakyansh-wav2vec2-tamil-tam-250
tags:
- generated_from_trainer
metrics:
- accuracy
model-index:
- name: vakyansh-wav2vec2-tamil-tam-250-audio-abuse-feature
  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. -->

# vakyansh-wav2vec2-tamil-tam-250-audio-abuse-feature

This model is a fine-tuned version of [Harveenchadha/vakyansh-wav2vec2-tamil-tam-250](https://huggingface.co/Harveenchadha/vakyansh-wav2vec2-tamil-tam-250) on the None dataset.
It achieves the following results on the evaluation set:
- Loss: 0.6061
- Accuracy: 0.7412
- Macro F1-score: 0.6531

## 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: 2e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- gradient_accumulation_steps: 4
- total_train_batch_size: 64
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 50

### Training results

| Training Loss | Epoch | Step | Validation Loss | Accuracy | Macro F1-score |
|:-------------:|:-----:|:----:|:---------------:|:--------:|:--------------:|
| 6.7458        | 0.77  | 10   | 6.7472          | 0.0      | 0.0            |
| 6.7056        | 1.54  | 20   | 6.6488          | 0.0      | 0.0            |
| 6.6158        | 2.31  | 30   | 6.5180          | 0.6307   | 0.0917         |
| 6.4651        | 3.08  | 40   | 6.2887          | 0.7143   | 0.4167         |
| 6.2508        | 3.85  | 50   | 5.9094          | 0.7197   | 0.4185         |
| 5.8959        | 4.62  | 60   | 5.5362          | 0.7197   | 0.4185         |
| 5.6179        | 5.38  | 70   | 5.2347          | 0.7197   | 0.4185         |
| 5.3048        | 6.15  | 80   | 4.9823          | 0.7197   | 0.4185         |
| 5.0858        | 6.92  | 90   | 4.7555          | 0.7197   | 0.4185         |
| 4.9195        | 7.69  | 100  | 4.5424          | 0.7197   | 0.4185         |
| 4.6747        | 8.46  | 110  | 4.3265          | 0.7197   | 0.4185         |
| 4.5861        | 9.23  | 120  | 4.1193          | 0.7197   | 0.4185         |
| 4.3397        | 10.0  | 130  | 3.9070          | 0.7197   | 0.4185         |
| 4.0926        | 10.77 | 140  | 3.6954          | 0.7197   | 0.4185         |
| 3.8859        | 11.54 | 150  | 3.4822          | 0.7197   | 0.4185         |
| 3.7254        | 12.31 | 160  | 3.2711          | 0.7197   | 0.4185         |
| 3.5303        | 13.08 | 170  | 3.0599          | 0.7197   | 0.4185         |
| 3.2531        | 13.85 | 180  | 2.8502          | 0.7197   | 0.4185         |
| 3.0184        | 14.62 | 190  | 2.6448          | 0.7197   | 0.4185         |
| 3.0006        | 15.38 | 200  | 2.4472          | 0.7197   | 0.4185         |
| 2.6674        | 16.15 | 210  | 2.2526          | 0.7197   | 0.4185         |
| 2.4455        | 16.92 | 220  | 2.0649          | 0.7197   | 0.4185         |
| 2.2702        | 17.69 | 230  | 1.8883          | 0.7197   | 0.4185         |
| 2.0536        | 18.46 | 240  | 1.7233          | 0.7197   | 0.4185         |
| 2.0643        | 19.23 | 250  | 1.5730          | 0.7197   | 0.4185         |
| 1.8006        | 20.0  | 260  | 1.4368          | 0.7197   | 0.4185         |
| 1.6975        | 20.77 | 270  | 1.3112          | 0.7197   | 0.4185         |
| 1.4407        | 21.54 | 280  | 1.2015          | 0.7197   | 0.4185         |
| 1.2971        | 22.31 | 290  | 1.1050          | 0.7197   | 0.4185         |
| 1.3202        | 23.08 | 300  | 1.0219          | 0.7197   | 0.4185         |
| 1.1292        | 23.85 | 310  | 0.9490          | 0.7197   | 0.4185         |
| 1.1055        | 24.62 | 320  | 0.8879          | 0.7197   | 0.4185         |
| 0.9817        | 25.38 | 330  | 0.8366          | 0.7197   | 0.4185         |
| 0.9296        | 26.15 | 340  | 0.7906          | 0.7197   | 0.4185         |
| 0.8306        | 26.92 | 350  | 0.7506          | 0.7197   | 0.4185         |
| 0.8303        | 27.69 | 360  | 0.7171          | 0.7197   | 0.4185         |
| 0.8421        | 28.46 | 370  | 0.6953          | 0.7197   | 0.4185         |
| 0.7964        | 29.23 | 380  | 0.6650          | 0.7197   | 0.4185         |
| 0.7528        | 30.0  | 390  | 0.6470          | 0.7197   | 0.4185         |
| 0.7305        | 30.77 | 400  | 0.6345          | 0.7197   | 0.4185         |
| 0.6702        | 31.54 | 410  | 0.6163          | 0.7385   | 0.4937         |
| 0.6416        | 32.31 | 420  | 0.6118          | 0.7547   | 0.5507         |
| 0.608         | 33.08 | 430  | 0.6086          | 0.7547   | 0.5507         |
| 0.6659        | 33.85 | 440  | 0.5981          | 0.7574   | 0.5949         |
| 0.5839        | 34.62 | 450  | 0.6068          | 0.7547   | 0.6570         |
| 0.6167        | 35.38 | 460  | 0.5894          | 0.7763   | 0.6479         |
| 0.5991        | 36.15 | 470  | 0.5947          | 0.7412   | 0.6531         |
| 0.5839        | 36.92 | 480  | 0.5938          | 0.7574   | 0.6771         |
| 0.5533        | 37.69 | 490  | 0.5922          | 0.7520   | 0.6399         |
| 0.4998        | 38.46 | 500  | 0.6203          | 0.7358   | 0.6625         |
| 0.5508        | 39.23 | 510  | 0.5865          | 0.7493   | 0.6278         |
| 0.5159        | 40.0  | 520  | 0.5963          | 0.7385   | 0.6670         |
| 0.5344        | 40.77 | 530  | 0.5946          | 0.7439   | 0.6420         |
| 0.5039        | 41.54 | 540  | 0.5979          | 0.7466   | 0.6526         |
| 0.5456        | 42.31 | 550  | 0.5999          | 0.7358   | 0.6707         |
| 0.4822        | 43.08 | 560  | 0.5845          | 0.7493   | 0.6437         |
| 0.4864        | 43.85 | 570  | 0.6035          | 0.7439   | 0.6779         |
| 0.4623        | 44.62 | 580  | 0.5961          | 0.7520   | 0.6519         |
| 0.475         | 45.38 | 590  | 0.6066          | 0.7439   | 0.6651         |
| 0.4887        | 46.15 | 600  | 0.6014          | 0.7466   | 0.6603         |
| 0.506         | 46.92 | 610  | 0.6012          | 0.7412   | 0.6604         |
| 0.5296        | 47.69 | 620  | 0.5986          | 0.7439   | 0.6503         |
| 0.5255        | 48.46 | 630  | 0.6003          | 0.7439   | 0.6503         |
| 0.4667        | 49.23 | 640  | 0.6038          | 0.7466   | 0.6553         |
| 0.4334        | 50.0  | 650  | 0.6061          | 0.7412   | 0.6531         |


### Framework versions

- Transformers 4.33.0
- Pytorch 2.0.0
- Datasets 2.1.0
- Tokenizers 0.13.3