SG风格数据集 33场对话200回复 为什么Num examples 只有30？

[Kas1o]

[INFO|trainer.py:641] 2024-06-23 21:33:50,843 >> Using auto half precision backend
[INFO|trainer.py:2078] 2024-06-23 21:33:51,146 >> ***** Running training *****
[INFO|trainer.py:2079] 2024-06-23 21:33:51,147 >>   Num examples = 30
[INFO|trainer.py:2080] 2024-06-23 21:33:51,147 >>   Num Epochs = 1
[INFO|trainer.py:2081] 2024-06-23 21:33:51,147 >>   Instantaneous batch size per device = 1
[INFO|trainer.py:2084] 2024-06-23 21:33:51,147 >>   Total train batch size (w. parallel, distributed & accumulation) = 4
[INFO|trainer.py:2085] 2024-06-23 21:33:51,147 >>   Gradient Accumulation steps = 4
[INFO|trainer.py:2086] 2024-06-23 21:33:51,148 >>   Total optimization steps = 7
[INFO|trainer.py:2087] 2024-06-23 21:33:51,159 >>   Number of trainable parameters = 27,394,04

几乎瞬间就调完了, 2080Ti应该不会这么快吧。我感觉哪里有问题？🤨

[Kas1o]

我不太清楚，我是应该为一场对话中的每个gpt回复都创建一个会话，还是只需要一个会话？