File size: 2,477 Bytes
259494f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
library_name: transformers
license: apache-2.0
base_model: openai/whisper-small
tags:
- generated_from_trainer
datasets:
- common_voice_22_0
metrics:
- wer
model-index:
- name: openai/whisper-small
  results:
  - task:
      name: Automatic Speech Recognition
      type: automatic-speech-recognition
    dataset:
      name: common_voice_22_0
      type: common_voice_22_0
      config: eu
      split: test
      args: eu
    metrics:
    - name: Wer
      type: wer
      value: 12.160220434106431
---

<!-- 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. -->

# openai/whisper-small

This model is a fine-tuned version of [openai/whisper-small](https://huggingface.co/openai/whisper-small) on the common_voice_22_0 dataset.
It achieves the following results on the evaluation set:
- Loss: 0.3891
- Wer: 12.1602

## 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: 3.75e-05
- train_batch_size: 64
- eval_batch_size: 32
- seed: 42
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- training_steps: 100000
- mixed_precision_training: Native AMP

### Training results

| Training Loss | Epoch    | Step  | Validation Loss | Wer     |
|:-------------:|:--------:|:-----:|:---------------:|:-------:|
| 0.0113        | 10.8225  | 5000  | 0.2778          | 12.1163 |
| 0.0047        | 21.6450  | 10000 | 0.3096          | 11.8923 |
| 0.0059        | 32.4675  | 15000 | 0.3285          | 12.1856 |
| 0.0028        | 43.2900  | 20000 | 0.3573          | 11.6886 |
| 0.0015        | 54.1126  | 25000 | 0.3549          | 11.6117 |
| 0.0028        | 64.9351  | 30000 | 0.3610          | 11.9312 |
| 0.0013        | 75.7576  | 35000 | 0.3711          | 11.6683 |
| 0.0009        | 86.5801  | 40000 | 0.3756          | 11.6176 |
| 0.0008        | 97.4026  | 45000 | 0.3801          | 11.8213 |
| 0.0009        | 108.2251 | 50000 | 0.3891          | 12.1602 |


### Framework versions

- Transformers 4.52.3
- Pytorch 2.6.0+cu124
- Datasets 3.6.0
- Tokenizers 0.21.1