🚀 ai-light-dance_chord_ft_wav2vec2-large-xlsr-53
本模型是 facebook/wav2vec2-large-xlsr-53 在 GARY109/AI_LIGHT_DANCE - ONSET-CHORD2 數據集上的微調版本。它在評估集上取得了以下結果:
- 損失值:1.8722
- 字錯率(Wer):0.9590
📚 詳細文檔
模型描述
更多信息待補充。
預期用途與限制
更多信息待補充。
訓練和評估數據
更多信息待補充。
🔧 技術細節
訓練過程
訓練超參數
訓練期間使用了以下超參數:
- 學習率(learning_rate):3e-05
- 訓練批次大小(train_batch_size):10
- 評估批次大小(eval_batch_size):10
- 隨機種子(seed):42
- 梯度累積步數(gradient_accumulation_steps):16
- 總訓練批次大小(total_train_batch_size):160
- 優化器(optimizer):Adam,β值為(0.9, 0.999),ε值為 1e-08
- 學習率調度器類型(lr_scheduler_type):線性
- 學習率調度器熱身步數(lr_scheduler_warmup_steps):100
- 訓練輪數(num_epochs):50.0
- 混合精度訓練(mixed_precision_training):原生自動混合精度(Native AMP)
訓練結果
訓練損失 |
輪數 |
步數 |
驗證損失 |
字錯率(Wer) |
5.1857 |
1.0 |
126 |
4.5913 |
1.0 |
3.0939 |
2.0 |
252 |
3.0160 |
1.0 |
2.8403 |
3.0 |
378 |
2.7337 |
1.0 |
2.2525 |
4.0 |
504 |
2.5588 |
0.9825 |
2.0291 |
5.0 |
630 |
2.5216 |
0.9701 |
1.9083 |
6.0 |
756 |
2.3990 |
0.9514 |
1.8745 |
7.0 |
882 |
2.2781 |
0.9474 |
1.8222 |
8.0 |
1008 |
2.2360 |
0.9471 |
1.7871 |
9.0 |
1134 |
2.1960 |
0.9463 |
1.7225 |
10.0 |
1260 |
2.0775 |
0.9464 |
1.6856 |
11.0 |
1386 |
2.0817 |
0.9518 |
1.6903 |
12.0 |
1512 |
2.0607 |
0.9534 |
1.6034 |
13.0 |
1638 |
1.9956 |
0.9504 |
1.6171 |
14.0 |
1764 |
2.0099 |
0.9490 |
1.5508 |
15.0 |
1890 |
2.0424 |
0.9591 |
1.539 |
16.0 |
2016 |
1.9728 |
0.9600 |
1.5176 |
17.0 |
2142 |
2.0421 |
0.9628 |
1.5088 |
18.0 |
2268 |
1.9428 |
0.9598 |
1.4739 |
19.0 |
2394 |
1.9886 |
0.9591 |
1.4228 |
20.0 |
2520 |
2.0164 |
0.9670 |
1.4277 |
21.0 |
2646 |
1.9968 |
0.9704 |
1.3834 |
22.0 |
2772 |
1.9882 |
0.9669 |
1.3768 |
23.0 |
2898 |
1.9519 |
0.9606 |
1.3747 |
24.0 |
3024 |
1.8923 |
0.9580 |
1.3533 |
25.0 |
3150 |
1.9767 |
0.9707 |
1.3312 |
26.0 |
3276 |
1.8993 |
0.9609 |
1.2743 |
27.0 |
3402 |
1.9494 |
0.9705 |
1.2924 |
28.0 |
3528 |
1.9019 |
0.9631 |
1.2621 |
29.0 |
3654 |
1.9110 |
0.9596 |
1.2387 |
30.0 |
3780 |
1.9118 |
0.9627 |
1.228 |
31.0 |
3906 |
1.8722 |
0.9590 |
1.1938 |
32.0 |
4032 |
1.8890 |
0.9599 |
1.1887 |
33.0 |
4158 |
1.9175 |
0.9653 |
1.1807 |
34.0 |
4284 |
1.8983 |
0.9649 |
1.1553 |
35.0 |
4410 |
1.9246 |
0.9703 |
1.1448 |
36.0 |
4536 |
1.9248 |
0.9705 |
1.1146 |
37.0 |
4662 |
1.9747 |
0.9804 |
1.1394 |
38.0 |
4788 |
1.9119 |
0.9723 |
1.1206 |
39.0 |
4914 |
1.8931 |
0.9630 |
1.0892 |
40.0 |
5040 |
1.9243 |
0.9668 |
1.104 |
41.0 |
5166 |
1.8965 |
0.9671 |
1.054 |
42.0 |
5292 |
1.9477 |
0.9755 |
1.0922 |
43.0 |
5418 |
1.8969 |
0.9699 |
1.0484 |
44.0 |
5544 |
1.9423 |
0.9733 |
1.0567 |
45.0 |
5670 |
1.9412 |
0.9745 |
1.0615 |
46.0 |
5796 |
1.9076 |
0.9674 |
1.0201 |
47.0 |
5922 |
1.9384 |
0.9743 |
1.0664 |
48.0 |
6048 |
1.9509 |
0.9816 |
1.0498 |
49.0 |
6174 |
1.9426 |
0.9757 |
1.0303 |
50.0 |
6300 |
1.9477 |
0.9781 |
框架版本
- Transformers 4.21.0.dev0
- Pytorch 1.9.1+cu102
- Datasets 2.3.3.dev0
- Tokenizers 0.12.1
📄 許可證
本模型採用 Apache-2.0 許可證。