完美解决：RuntimeError: CUDA error: CUBLAS_STATUS_INTERNAL_ERROR when calling `cublasCreate(handle)

「爱情、让人受尽委屈。」 2022-11-12 09:34 1697阅读 0赞

报错场景：

self.scores = nn.ParameterList()
            self.scores.append(Parameter(torch.FloatTensor(nfeat, 1)))
            for i in range(nhidlayer):
                self.scores.append(Parameter(torch.FloatTensor(nhid, 1)))
    
    ## 这一句报错，也就是做矩阵乘法@时报错
    s_i = torch.sigmoid(fea @ self.scores[0] + self.bias[0])

我在网上搜了很多，都是啥embedding的原因，明显跟我身处的场景不一样

[https://blog.csdn.net/xiaoxiaowantong/article/details/109689146][https_blog.csdn.net_xiaoxiaowantong_article_details_109689146]

之后，我再stackflow找到了答案，[https://stackoverflow.com/questions/66567324/pytorch-error-cuda-error-cublas-status-internal-error-when-calling-cublascrea][https_stackoverflow.com_questions_66567324_pytorch-error-cuda-error-cublas-status-internal-error-when-calling-cublascrea]

> 就是pytorch的版本有问题，我之前用的版本是1.8.0，按照解决办法，换成1.7.1的。
> 
> pip uninstall torch
> 
> pip install torch==1.7.1

[https_blog.csdn.net_xiaoxiaowantong_article_details_109689146]: https://blog.csdn.net/xiaoxiaowantong/article/details/109689146
[https_stackoverflow.com_questions_66567324_pytorch-error-cuda-error-cublas-status-internal-error-when-calling-cublascrea]: https://stackoverflow.com/questions/66567324/pytorch-error-cuda-error-cublas-status-internal-error-when-calling-cublascrea