How to generate/get data for "alignment_dir" ?

[vetmax7]

Hello everyone!

I want to try train openfold model on small amount of data for understanding the process. How can I generate alignment_dir? Or where can I find ready data?

According to ReadMe, there are several ways: alphafold method, mmseq2, RODA and ProteinNet. But they all take a long time and require a lot of GPU resources.

I tried mmseq2 but I've got an error about memory. And I downloaded a part of RODA, applied a script: "flatten_roda" But I've got an error RuntimeError: CUDA error: CUBLAS_STATUS_INVALID_VALUE when calling cublasGemmEx( handle, opa, opb, m, n, k, &falpha, a, CUDA_R_16BF, lda, b, CUDA_R_16BF, ldb, &fbeta, c, CUDA_R_16BF, ldc, CUDA_R_32F, CUBLAS_GEMM_DFALT_TENSOR_OP)

What is a simple way for a really small data set? Is it possible?