I trained a DeepVariant model on WGS/Illumina data (input shape: [100, 221, 6]). When I try to run it directly on PACBIO or ONT reads, I get the following fatal error: I0918 07:32:49.111899 ...

When running tests/test_bmm_fp8.py with cuda 13.0 B200, we encounter an Segfault. We can pass the test with cuda 12.9. We get the first error for test case: False ...