File size: 3,341 Bytes
1004f4a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5cdba00
 
1004f4a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5cdba00
 
 
1004f4a
 
5cdba00
 
1004f4a
 
 
5cdba00
1004f4a
 
 
 
 
 
5cdba00
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1004f4a
 
 
 
 
 
 
 
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
96
---
license: apache-2.0
tags:
- generated_from_trainer
datasets:
- marsyas/gtzan
metrics:
- accuracy
model-index:
- name: distilhubert-finetuned-gtzan
  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. -->

# 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: 1.2658
- Accuracy: 0.8

## 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: 4
- eval_batch_size: 4
- seed: 42
- distributed_type: multi-GPU
- gradient_accumulation_steps: 2
- total_train_batch_size: 8
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 30
- mixed_precision_training: Native AMP

### Training results

| Training Loss | Epoch | Step | Validation Loss | Accuracy |
|:-------------:|:-----:|:----:|:---------------:|:--------:|
| 2.1071        | 1.0   | 112  | 2.1453          | 0.33     |
| 1.6165        | 2.0   | 225  | 1.6129          | 0.59     |
| 1.2842        | 3.0   | 337  | 1.2084          | 0.68     |
| 0.9805        | 4.0   | 450  | 0.8842          | 0.74     |
| 0.5216        | 5.0   | 562  | 0.7350          | 0.78     |
| 0.5017        | 6.0   | 675  | 0.8196          | 0.77     |
| 0.1998        | 7.0   | 787  | 0.6709          | 0.8      |
| 0.3662        | 8.0   | 900  | 0.8483          | 0.78     |
| 0.2711        | 9.0   | 1012 | 0.8567          | 0.81     |
| 0.0183        | 10.0  | 1125 | 0.8994          | 0.82     |
| 0.0299        | 11.0  | 1237 | 1.2142          | 0.8      |
| 0.0064        | 12.0  | 1350 | 1.0208          | 0.81     |
| 0.004         | 13.0  | 1462 | 1.0619          | 0.81     |
| 0.0031        | 14.0  | 1575 | 1.1454          | 0.79     |
| 0.0028        | 15.0  | 1687 | 1.1010          | 0.81     |
| 0.0023        | 16.0  | 1800 | 1.0595          | 0.8      |
| 0.0017        | 17.0  | 1912 | 1.1340          | 0.8      |
| 0.0015        | 18.0  | 2025 | 1.1760          | 0.81     |
| 0.0014        | 19.0  | 2137 | 1.1361          | 0.81     |
| 0.0012        | 20.0  | 2250 | 1.2138          | 0.81     |
| 0.0011        | 21.0  | 2362 | 1.1366          | 0.81     |
| 0.0012        | 22.0  | 2475 | 1.1662          | 0.8      |
| 0.0011        | 23.0  | 2587 | 1.1491          | 0.8      |
| 0.0009        | 24.0  | 2700 | 1.1287          | 0.81     |
| 0.0009        | 25.0  | 2812 | 1.2027          | 0.81     |
| 0.0009        | 26.0  | 2925 | 1.1740          | 0.81     |
| 0.0009        | 27.0  | 3037 | 1.2011          | 0.81     |
| 0.0009        | 28.0  | 3150 | 1.2523          | 0.8      |
| 0.0008        | 29.0  | 3262 | 1.2494          | 0.81     |
| 0.0007        | 29.87 | 3360 | 1.2658          | 0.8      |


### Framework versions

- Transformers 4.29.2
- Pytorch 1.13.1+cu117
- Datasets 2.12.0
- Tokenizers 0.13.3