File size: 2,558 Bytes
f353149
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
base_model: openai/whisper-medium
datasets:
- facebook/voxpopuli
language:
- it
library_name: peft
license: apache-2.0
metrics:
- wer
tags:
- generated_from_trainer
model-index:
- name: Whisper Medium
  results:
  - task:
      type: automatic-speech-recognition
      name: Automatic Speech Recognition
    dataset:
      name: facebook/voxpopuli
      type: facebook/voxpopuli
      config: default
      split: None
      args: default
    metrics:
    - type: wer
      value: 10.9375
      name: Wer
---

<!-- 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 Medium

This model is a fine-tuned version of [openai/whisper-medium](https://huggingface.co/openai/whisper-medium) on the facebook/voxpopuli dataset.
It achieves the following results on the evaluation set:
- Loss: 0.4874
- Wer: 10.9375

## 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: 0.0001
- train_batch_size: 32
- eval_batch_size: 16
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 64
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 100
- training_steps: 1200
- mixed_precision_training: Native AMP

### Training results

| Training Loss | Epoch  | Step | Validation Loss | Wer     |
|:-------------:|:------:|:----:|:---------------:|:-------:|
| 2.2174        | 0.5714 | 100  | 1.9102          | 49.4792 |
| 0.2353        | 1.1429 | 200  | 0.3485          | 30.7292 |
| 0.1668        | 1.7143 | 300  | 0.7634          | 21.875  |
| 0.118         | 2.2857 | 400  | 0.6914          | 11.9792 |
| 0.0931        | 2.8571 | 500  | 0.5523          | 15.1042 |
| 0.0851        | 3.4286 | 600  | 0.6818          | 13.0208 |
| 0.0751        | 4.0    | 700  | 0.6348          | 11.9792 |
| 0.066         | 4.5714 | 800  | 0.6576          | 11.9792 |
| 0.0604        | 5.1429 | 900  | 0.4125          | 10.9375 |
| 0.0564        | 5.7143 | 1000 | 0.6815          | 10.9375 |
| 0.0499        | 6.2857 | 1100 | 0.4861          | 11.4583 |
| 0.0472        | 6.8571 | 1200 | 0.4874          | 10.9375 |


### Framework versions

- PEFT 0.12.0
- Transformers 4.43.1
- Pytorch 2.4.1+cu121
- Datasets 3.0.0
- Tokenizers 0.19.1