一段时间以来，我一直在bert上训练自定义数据集。在训练过程中，我观察到，尽管训练损失在减少，val精度仍保持为0.5。我已经尝试了Huggingface中所有可用的所有BERT模型，但不知道该怎么做。这是训练期间的输出：
step: 0 of total steps: 250 step: 25 of total steps: 250 step: 50 of total steps: 250 step: 75 of total steps: 250 step: 100 of total steps: 250 step: 125 of total steps: 250 step: 150 of total steps: 250 step: 175 of total steps: 250 step: 200 of total steps: 250 step: 225 of total steps: 250 Average loss for epoch: 0.4582762689590454 /usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:6: DeprecationWarning: elementwise comparison failed; this will raise an error in the future. Accuracy: 0.50 Eval loss: 0.40 Epoch: 1 step: 0 of total steps: 250 step: 25 of total steps: 250 step: 50 of total steps: 250 step: 75 of total steps: 250 step: 100 of total steps: 250 step: 125 of total steps: 250 step: 150 of total steps: 250 step: 175 of total steps: 250 step: 200 of total steps: 250 step: 225 of total steps: 250 Average loss for epoch: 0.3786401364207268 Accuracy: 0.50 Eval loss: 0.39``` Can this be overfitting by any chance?