File size: 2,848 Bytes
e94d61e
a276450
 
e94d61e
 
a276450
e94d61e
 
a276450
e94d61e
 
24a784e
e94d61e
a276450
e94d61e
 
 
24a784e
e94d61e
a276450
 
e94d61e
 
 
 
24a784e
a276450
24a784e
e94d61e
 
a276450
e94d61e
a276450
e94d61e
a276450
 
e94d61e
9aefc3a
 
e94d61e
 
 
 
 
 
b43bcdc
e94d61e
 
 
 
 
 
 
 
 
 
c005abf
257c7d5
 
e94d61e
 
 
 
b43bcdc
e94d61e
 
 
 
 
 
c005abf
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e94d61e
 
 
 
 
 
 
 
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
---
language:
- eu
license: apache-2.0
tags:
- whisper-event
- generated_from_trainer
datasets:
- mozilla-foundation/common_voice_13_0
metrics:
- wer
base_model: openai/whisper-medium
model-index:
- name: Whisper Small Basque
  results:
  - task:
      type: automatic-speech-recognition
      name: Automatic Speech Recognition
    dataset:
      name: mozilla-foundation/common_voice_13_0 eu
      type: mozilla-foundation/common_voice_13_0
      config: eu
      split: test
      args: eu
    metrics:
    - type: wer
      value: 12.839726193851513
      name: Wer
---

# Whisper Small Basque

This model is a fine-tuned version of [openai/whisper-medium](https://huggingface.co/openai/whisper-medium) on the mozilla-foundation/common_voice_13_0 eu dataset.
It achieves the following results on the evaluation set:
- Loss: 0.2287
- Wer: 12.8397

If you need to use this model with [whisper.cpp](https://github.com/ggerganov/whisper.cpp), you can download the ggml file: [ggml-medium-eu.bin](https://huggingface.co/xezpeleta/whisper-medium-eu/blob/main/ggml-medium.eu.bin)

## 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: 4
- 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: 500
- training_steps: 8000
- mixed_precision_training: Native AMP

### Training results

| Training Loss | Epoch | Step | Validation Loss | Wer     |
|:-------------:|:-----:|:----:|:---------------:|:-------:|
| 0.4415        | 0.06  | 500  | 0.5092          | 36.9699 |
| 0.4206        | 0.12  | 1000 | 0.4144          | 28.3365 |
| 0.272         | 0.19  | 1500 | 0.3554          | 24.7438 |
| 0.2681        | 0.25  | 2000 | 0.3271          | 22.1414 |
| 0.2099        | 0.31  | 2500 | 0.2973          | 19.5350 |
| 0.2283        | 0.38  | 3000 | 0.2760          | 18.5042 |
| 0.1477        | 1.03  | 3500 | 0.2637          | 17.1493 |
| 0.1008        | 1.09  | 4000 | 0.2592          | 16.3939 |
| 0.0866        | 1.15  | 4500 | 0.2561          | 15.8066 |
| 0.0915        | 1.21  | 5000 | 0.2411          | 15.0310 |
| 0.0803        | 1.28  | 5500 | 0.2330          | 14.7616 |
| 0.0674        | 1.34  | 6000 | 0.2325          | 13.8462 |
| 0.0679        | 1.4   | 6500 | 0.2299          | 13.5809 |
| 0.027         | 2.05  | 7000 | 0.2304          | 13.3805 |
| 0.0231        | 2.11  | 7500 | 0.2287          | 12.8397 |
| 0.0285        | 2.18  | 8000 | 0.2304          | 12.8883 |


### Framework versions

- Transformers 4.26.0.dev0
- Pytorch 1.13.1+cu117
- Datasets 2.8.1.dev0
- Tokenizers 0.13.2