RuntimeError: CUDA error: CUBLAS_STATUS_EXECUTION_FAILED when calling cublasGemmEx( handle, opa, opb, m, n, k, &falpha, a, CUDA_R_16F, lda, b, CUDA_R_16F, ldb, &fbeta, c, CUDA_R_16F, ldc, CUDA_R_32F, CUBLAS_GEMM_DFALT_TENSOR_OP)

[LukeKnutson9]

I found an error
RuntimeError: CUDA error: CUBLAS_STATUS_EXECUTION_FAILED when calling cublasGemmEx( handle, opa, opb, m, n, k, &falpha, a, CUDA_R_16F, lda, b, CUDA_R_16F, ldb, &fbeta, c, CUDA_R_16F, ldc, CUDA_R_32F, CUBLAS_GEMM_DFALT_TENSOR_OP)``
when attempting to call the run the sample code provided in the ReadMe,:

`from big_sleep import Imagine

dream = Imagine(
    text = "fire in the sky",
    lr = 5e-2,
    save_every = 50,
    save_progress = True
)

dream()`

The error seems to have occurred here:
File "D:\Anaconda\lib\site-packages\torch\nn\functional.py", line 1753, in linear return torch._C._nn.linear(input, weight, bias)

Does anyone know why this would occur? I am running my code on a 64 bit Windows 10 laptop with a GeForce GTX 1660 Ti Graphics Card

[apprehensivetarget]

I am gettind the same error.

[TheGildedGoo]

Having the same error utilizing a 1660 Super. Thinking it may have something to do with the GPU.

[apprehensivetarget]

I fixed it by reinstalling pytotch

[Rerumu]

Having this issue on Debian with my GTX 1660 Super. Reinstalling pytorch did not help.
After some testing I found it may be a memory issue. Setting cutout size to 17 allows it to run, but any higher produces the error.

[anvarganiev]

I had wrong tensor shape in NN architecture. In one place there was 32x0x6 tensor.

[BassemKaroui]

I had the same error, and thx to @anvarganiev I noticed that one tensor had the wrong shape

[0xBYTESHIFT]

Got this issue too, using the default CLI, not a .py file with my parameters. How to fix it?