🚀 ModerationGPT
ModerationGPT是基於t5-small在None數據集上微調得到的模型。該模型在評估集上取得了出色的成績,能為相關任務提供高效準確的支持。
🚀 快速開始
此模型是 t5-small 在 None 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(Loss):0.2177
- Rouge1:81.7742
- Rouge2:77.8168
- Rougel:81.7812
- Rougelsum:81.7593
- 生成長度(Gen Len):13.6851
🔧 技術細節
訓練超參數
訓練過程中使用了以下超參數:
- 學習率(learning_rate):2e-05
- 訓練批次大小(train_batch_size):128
- 評估批次大小(eval_batch_size):128
- 隨機種子(seed):42
- 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
- 訓練輪數(num_epochs):20
訓練結果
訓練損失 |
輪數 |
步數 |
驗證損失 |
Rouge1 |
Rouge2 |
Rougel |
Rougelsum |
生成長度 |
無日誌記錄 |
1.0 |
335 |
1.0577 |
54.2885 |
42.6805 |
54.1114 |
54.2064 |
15.8069 |
1.4482 |
2.0 |
670 |
0.5655 |
71.3866 |
65.6295 |
71.3181 |
71.3413 |
13.3214 |
0.7111 |
3.0 |
1005 |
0.4278 |
73.535 |
68.2492 |
73.5214 |
73.5207 |
13.3682 |
0.7111 |
4.0 |
1340 |
0.3626 |
74.1375 |
69.0355 |
74.1302 |
74.1116 |
13.409 |
0.508 |
5.0 |
1675 |
0.3200 |
75.8217 |
69.4616 |
75.8126 |
75.7987 |
13.4467 |
0.4281 |
6.0 |
2010 |
0.2923 |
80.1152 |
75.0253 |
80.1019 |
80.0814 |
13.5971 |
0.4281 |
7.0 |
2345 |
0.2740 |
81.1188 |
76.4582 |
81.1144 |
81.0866 |
13.6413 |
0.3818 |
8.0 |
2680 |
0.2621 |
81.4117 |
76.9557 |
81.4064 |
81.3814 |
13.6591 |
0.3505 |
9.0 |
3015 |
0.2521 |
81.5234 |
77.2074 |
81.5176 |
81.4917 |
13.6683 |
0.3505 |
10.0 |
3350 |
0.2461 |
81.6226 |
77.3821 |
81.6243 |
81.5999 |
13.6729 |
0.3331 |
11.0 |
3685 |
0.2401 |
81.6549 |
77.4807 |
81.6601 |
81.6397 |
13.6769 |
0.3176 |
12.0 |
4020 |
0.2341 |
81.7157 |
77.6237 |
81.7192 |
81.6996 |
13.6807 |
0.3176 |
13.0 |
4355 |
0.2314 |
81.7288 |
77.6693 |
81.7346 |
81.7125 |
13.6814 |
0.3068 |
14.0 |
4690 |
0.2271 |
81.7534 |
77.7293 |
81.758 |
81.7446 |
13.6834 |
0.3011 |
15.0 |
5025 |
0.2239 |
81.7569 |
77.7588 |
81.7643 |
81.7426 |
13.6839 |
0.3011 |
16.0 |
5360 |
0.2209 |
81.761 |
77.7766 |
81.7681 |
81.7474 |
13.6855 |
0.2951 |
17.0 |
5695 |
0.2201 |
81.7483 |
77.7596 |
81.7516 |
81.7331 |
13.6847 |
0.2904 |
18.0 |
6030 |
0.2188 |
81.7582 |
77.7865 |
81.7605 |
81.7428 |
13.6842 |
0.2904 |
19.0 |
6365 |
0.2179 |
81.7842 |
77.8249 |
81.7899 |
81.7721 |
13.686 |
0.2893 |
20.0 |
6700 |
0.2177 |
81.7742 |
77.8168 |
81.7812 |
81.7593 |
13.6851 |
框架版本
- Transformers 4.29.2
- Pytorch 2.0.1+cu117
- Datasets 2.12.0
- Tokenizers 0.13.3
📄 許可證
本模型採用 Apache-2.0 許可證。