运行train.py时报错

我在准备好自己的数据集之后 基于现有的secgpt进行二次训练 再运行 train.py 时 报错 ：RuntimeError: CUDA error: CUBLAS_STATUS_NOT_INITIALIZED when calling `cublasCreate(handle)` 请问这是什么原因导致的？ 当初训练secgpt遇到同样的错误您是如何解决的。 期待您的解答 谢谢！

Start with train.py and reproduce the reported CUBLAS_STATUS_NOT_INITIALIZED error using the reporter's prepared dataset. Inspect the training entry point and the CUDA initialization path; done means identifying a reproducible cause and documenting a verified resolution.

Written by the indexing model from the issue text.