File size: 3,041 Bytes
a806551
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
language:
- en
license: apache-2.0
base_model: piyushmaharana/outcomes-whisper-tiny-v1
tags:
- generated_from_trainer
datasets:
- ray-outcomes-ai/big-transcript-pronounce
metrics:
- wer
model-index:
- name: OutcomesAI-Whisper-tiny-v1.2
  results:
  - task:
      name: Automatic Speech Recognition
      type: automatic-speech-recognition
    dataset:
      name: big-transcript-pronounce
      type: ray-outcomes-ai/big-transcript-pronounce
      args: 'config: en, split: test'
    metrics:
    - name: Wer
      type: wer
      value: 2.8199566160520604
---

<!-- 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. -->

# OutcomesAI-Whisper-tiny-v1.2

This model is a fine-tuned version of [piyushmaharana/outcomes-whisper-tiny-v1](https://huggingface.co/piyushmaharana/outcomes-whisper-tiny-v1) on the big-transcript-pronounce dataset.
It achieves the following results on the evaluation set:
- Loss: 0.0602
- Wer: 2.8200

## 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: 1e-05
- train_batch_size: 16
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 100
- training_steps: 2000
- mixed_precision_training: Native AMP

### Training results

| Training Loss | Epoch | Step | Validation Loss | Wer    |
|:-------------:|:-----:|:----:|:---------------:|:------:|
| 0.0401        | 12.5  | 100  | 0.0998          | 5.4230 |
| 0.0006        | 25.0  | 200  | 0.0777          | 6.5076 |
| 0.0003        | 37.5  | 300  | 0.0723          | 4.1215 |
| 0.0002        | 50.0  | 400  | 0.0691          | 3.4707 |
| 0.0001        | 62.5  | 500  | 0.0669          | 3.2538 |
| 0.0001        | 75.0  | 600  | 0.0656          | 3.0369 |
| 0.0001        | 87.5  | 700  | 0.0646          | 3.2538 |
| 0.0001        | 100.0 | 800  | 0.0635          | 3.4707 |
| 0.0001        | 112.5 | 900  | 0.0628          | 3.4707 |
| 0.0001        | 125.0 | 1000 | 0.0624          | 3.4707 |
| 0.0001        | 137.5 | 1100 | 0.0619          | 3.4707 |
| 0.0001        | 150.0 | 1200 | 0.0614          | 2.8200 |
| 0.0001        | 162.5 | 1300 | 0.0613          | 3.2538 |
| 0.0           | 175.0 | 1400 | 0.0609          | 3.2538 |
| 0.0           | 187.5 | 1500 | 0.0607          | 3.2538 |
| 0.0           | 200.0 | 1600 | 0.0606          | 3.2538 |
| 0.0           | 212.5 | 1700 | 0.0604          | 3.2538 |
| 0.0           | 225.0 | 1800 | 0.0605          | 3.4707 |
| 0.0           | 237.5 | 1900 | 0.0603          | 3.2538 |
| 0.0           | 250.0 | 2000 | 0.0602          | 2.8200 |


### Framework versions

- Transformers 4.44.0
- Pytorch 2.4.0+cu121
- Datasets 2.20.0
- Tokenizers 0.19.1