File size: 1,679 Bytes
c17bbff
 
f3b6341
c17bbff
 
2c013c0
 
c17bbff
b7340ea
f3b6341
c17bbff
 
 
 
 
f3b6341
c17bbff
b7340ea
2c013c0
f3b6341
 
c17bbff
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f3b6341
c17bbff
 
f3b6341
c17bbff
 
 
f3b6341
c17bbff
2c013c0
 
f3b6341
 
 
 
 
 
 
2c013c0
 
c17bbff
 
f3b6341
 
 
 
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
---
license: apache-2.0
base_model: openai/whisper-small
tags:
- generated_from_trainer
metrics:
- wer
model-index:
- name: whisper-small-ckb
  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. -->

# whisper-small-sorani

This model is a fine-tuned version of [openai/whisper-small](https://huggingface.co/openai/whisper-small) on the common voice v15 dataset.
It achieves the following results on the evaluation set:
- Loss: 0.0334
- Wer: 0.0613

## 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: 56
- eval_batch_size: 32
- seed: 42
- distributed_type: multi-GPU
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- training_steps: 5000

### Training results

| Training Loss | Epoch | Step | Validation Loss | Wer    |
|:-------------:|:-----:|:----:|:---------------:|:------:|
| 0.1005        | 0.55  | 1000 | 0.1405          | 0.2743 |
| 0.0711        | 1.09  | 2000 | 0.0858          | 0.1772 |
| 0.0609        | 1.64  | 3000 | 0.0585          | 0.1151 |
| 0.02          | 2.19  | 4000 | 0.0408          | 0.0789 |
| 0.0169        | 2.74  | 5000 | 0.0334          | 0.0613 |


### Framework versions

- Transformers 4.34.0.dev0
- Pytorch 2.0.1+cu117
- Datasets 2.14.5
- Tokenizers 0.14.0