File size: 2,574 Bytes
ae0d22c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
de44280
ae0d22c
 
 
 
 
 
 
 
 
de44280
 
ae0d22c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
de44280
 
 
ae0d22c
 
 
de44280
 
ae0d22c
 
 
 
 
de44280
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ae0d22c
 
 
 
de44280
ae0d22c
 
 
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
---
license: apache-2.0
base_model: ntu-spml/distilhubert
tags:
- generated_from_trainer
datasets:
- marsyas/gtzan
metrics:
- accuracy
model-index:
- name: distilhubert-finetuned-gtzan
  results:
  - task:
      name: Audio Classification
      type: audio-classification
    dataset:
      name: GTZAN
      type: marsyas/gtzan
      config: all
      split: train
      args: all
    metrics:
    - name: Accuracy
      type: accuracy
      value: 0.9
---

<!-- 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. -->

# distilhubert-finetuned-gtzan

This model is a fine-tuned version of [ntu-spml/distilhubert](https://huggingface.co/ntu-spml/distilhubert) on the GTZAN dataset.
It achieves the following results on the evaluation set:
- Loss: 0.6062
- Accuracy: 0.9

## 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: 5e-05
- train_batch_size: 8
- 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_ratio: 0.1
- num_epochs: 15

### Training results

| Training Loss | Epoch | Step | Validation Loss | Accuracy |
|:-------------:|:-----:|:----:|:---------------:|:--------:|
| 2.0776        | 1.0   | 113  | 1.9082          | 0.48     |
| 1.3768        | 2.0   | 226  | 1.3052          | 0.63     |
| 1.0741        | 3.0   | 339  | 0.9721          | 0.79     |
| 0.778         | 4.0   | 452  | 0.8452          | 0.76     |
| 0.6383        | 5.0   | 565  | 0.5935          | 0.85     |
| 0.3313        | 6.0   | 678  | 0.5947          | 0.81     |
| 0.3514        | 7.0   | 791  | 0.6064          | 0.8      |
| 0.0922        | 8.0   | 904  | 0.5759          | 0.81     |
| 0.1757        | 9.0   | 1017 | 0.4683          | 0.88     |
| 0.0496        | 10.0  | 1130 | 0.5958          | 0.86     |
| 0.0141        | 11.0  | 1243 | 0.5512          | 0.87     |
| 0.0345        | 12.0  | 1356 | 0.6297          | 0.86     |
| 0.0079        | 13.0  | 1469 | 0.6009          | 0.89     |
| 0.0072        | 14.0  | 1582 | 0.6069          | 0.9      |
| 0.007         | 15.0  | 1695 | 0.6062          | 0.9      |


### Framework versions

- Transformers 4.35.0.dev0
- Pytorch 2.0.1+cu118
- Datasets 2.14.5
- Tokenizers 0.14.0