File size: 4,089 Bytes
1004f4a 2496c05 1004f4a f28c2df 1004f4a f28c2df 1004f4a 5cdba00 1004f4a 124c8c1 1004f4a 124c8c1 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 97 98 99 |
---
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 on best epoch:
- Loss: 0.7305
- Accuracy: 0.9
## Model description
Distilhubert is distilled version of the [HuBERT](https://huggingface.co/docs/transformers/model_doc/hubert) and pretrained on data set with 16k frequency. <br/>
Architecture of this model is CTC or Connectionist Temporal Classification is a technique that is used with encoder-only transformer. <br/>
## Training and evaluation data
Training + Evaluation data set is GTZAN which is a popular dataset of 999 songs for music genre classification. <br/>
Each song is a 30-second clip from one of 10 genres of music, spanning disco to metal.<br/>
Train set is 899 songs and Evaluation set is 100 songs remainings.
## 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
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 35
- mixed_precision_training: Native AMP
### Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy |
|:-------------:|:-----:|:----:|:---------------:|:--------:|
| 2.1728 | 1.0 | 225 | 2.0896 | 0.42 |
| 1.4211 | 2.0 | 450 | 1.4951 | 0.55 |
| 1.2155 | 3.0 | 675 | 1.0669 | 0.72 |
| 1.0175 | 4.0 | 900 | 0.8862 | 0.69 |
| 0.3516 | 5.0 | 1125 | 0.6265 | 0.83 |
| 0.6135 | 6.0 | 1350 | 0.6485 | 0.78 |
| 0.0807 | 7.0 | 1575 | 0.6567 | 0.78 |
| 0.0303 | 8.0 | 1800 | 0.7615 | 0.83 |
| 0.2663 | 9.0 | 2025 | 0.6612 | 0.86 |
| 0.0026 | 10.0 | 2250 | 0.8354 | 0.85 |
| 0.0337 | 11.0 | 2475 | 0.6768 | 0.87 |
| 0.0013 | 12.0 | 2700 | 0.7718 | 0.87 |
| 0.001 | 13.0 | 2925 | 0.7570 | 0.88 |
| 0.0008 | 14.0 | 3150 | 0.8170 | 0.89 |
| 0.0006 | 15.0 | 3375 | 0.7920 | 0.89 |
| 0.0005 | 16.0 | 3600 | 0.9859 | 0.83 |
| 0.0004 | 17.0 | 3825 | 0.8190 | 0.9 |
| 0.0003 | 18.0 | 4050 | 0.7305 | 0.9 |
| 0.0003 | 19.0 | 4275 | 0.8025 | 0.88 |
| 0.0002 | 20.0 | 4500 | 0.8208 | 0.87 |
| 0.0003 | 21.0 | 4725 | 0.7358 | 0.88 |
| 0.0002 | 22.0 | 4950 | 0.8681 | 0.87 |
| 0.0002 | 23.0 | 5175 | 0.7831 | 0.9 |
| 0.0003 | 24.0 | 5400 | 0.8583 | 0.88 |
| 0.0002 | 25.0 | 5625 | 0.8138 | 0.88 |
| 0.0002 | 26.0 | 5850 | 0.7871 | 0.89 |
| 0.0002 | 27.0 | 6075 | 0.8893 | 0.88 |
| 0.0002 | 28.0 | 6300 | 0.8284 | 0.89 |
| 0.0001 | 29.0 | 6525 | 0.8388 | 0.89 |
| 0.0001 | 30.0 | 6750 | 0.8305 | 0.9 |
| 0.0001 | 31.0 | 6975 | 0.8377 | 0.88 |
| 0.0153 | 32.0 | 7200 | 0.8496 | 0.88 |
| 0.0001 | 33.0 | 7425 | 0.8381 | 0.88 |
| 0.0001 | 34.0 | 7650 | 0.8440 | 0.88 |
| 0.0001 | 35.0 | 7875 | 0.8458 | 0.88 |
### Framework versions
- Transformers 4.29.2
- Pytorch 1.13.1+cu117
- Datasets 2.12.0
- Tokenizers 0.13.3
|