[BUG/Help] 如何在 ptuning/main.py 中精确指定 EPOCH

[wizardforcel]

Is there an existing issue for this?

• [X] I have searched the existing issues

Current Behavior

[INFO|trainer.py:567] 2023-09-11 16:25:22,233 >> max_steps is given, it will override any value given in num_train_epochs
[INFO|trainer.py:1712] 2023-09-11 16:25:22,360 >> ***** Running training *****
[INFO|trainer.py:1713] 2023-09-11 16:25:22,360 >>   Num examples = 731
[INFO|trainer.py:1714] 2023-09-11 16:25:22,360 >>   Num Epochs = 9
[INFO|trainer.py:1715] 2023-09-11 16:25:22,360 >>   Instantaneous batch size per device = 1
[INFO|trainer.py:1718] 2023-09-11 16:25:22,360 >>   Total train batch size (w. parallel, distributed & accumulation) = 16
[INFO|trainer.py:1719] 2023-09-11 16:25:22,360 >>   Gradient Accumulation steps = 16
[INFO|trainer.py:1720] 2023-09-11 16:25:22,360 >>   Total optimization steps = 400
[INFO|trainer.py:1721] 2023-09-11 16:25:22,361 >>   Number of trainable parameters = 1,835,008

Expected Behavior

我的数据集大小是 731 条，批量大小是 1，然后 max_step 设置为 400，结果发现 n_epoch 为 9，这个是怎么计算的？

如果我想直接调整 Epoch 数量，应该怎么办？

Steps To Reproduce

.

Environment

- OS:
- Python:
- Transformers:
- PyTorch:
- CUDA Support (`python -c "import torch; print(torch.cuda.is_available())"`) :

Anything else?

No response