运行时错误Pytorch：at :: cuda :: blas :: gemm <float>参数ldb必须为正

Question

我为BO使用Ax，最终通过Pytorch通过Botorch和GPytorch。在某个时候，调用loss.backward()时会出现以下错误：

  File ".../model.py", line 496, in train_loop
    loss.backward()
  File ".../torch/tensor.py", line 166, in backward
    torch.autograd.backward(self, gradient, retain_graph, create_graph)
  File ".../torch/autograd/__init__.py", line 99, in backward
    allow_unreachable=True)  # allow_unreachable flag
RuntimeError: at::cuda::blas::gemm<float> argument ldb must be positive and less than 2147483647 but got 0

查看原始文件CUDABlass.cpp，该错误对应于正整数检查：

#define CUDABLAS_POSINT_CHECK(FD, X)         \
  TORCH_CHECK(                               \
      (X > 0 && X <= INT_MAX),               \
      "at::cuda::blas::" #FD " argument " #X \
      " must be positive and less than ",    \
      INT_MAX,                               \
      " but got ",                           \
      X)

我不知道错误的来源。我怀疑有一些Search Space变量，但我看不到它们有任何错误（或者相同，问题与Ax，Botorch或Gpytorch无关）。有人遇到过此错误吗？你怎么解决的？