Get tensor-> data when the tensor's backend is CUDA

[MirkoDeVita98]

Hello! I am trying to read the data content of a tensor that is supposed to be on GPU, when I try to access the content of data normally I get segmentation fault due to accessing an address that is not on RAM. It all comes to making this to work, but I am missing where I should get the backend (first argument of ggml_backend_tensor_get_async) from:

std::vector<int32_t> dst_data(ggml_nelements(dst));
ggml_backend_tensor_get_async(backend,dst, dst_data.data(), 0, ggml_nbytes(dst));

Thank you in advance for the help!

[slaren]

You can use ggml_backend_tensor_get instead.