Ab initio reconstruction broken?

Dear all,

After running an local motion correction (1200px box size), we evaluated two different Ab initio reconstructions of particles - without symmetry imposed and with symmetry (I2). As a comparison, we also ran both on particles before local motion correction.
While the job for C1 reconstruction goes through in approx. 6 hours for 21k particles, the job for I2 reconstruction is projected to take an unreasonably long time. Time per iteration increases from 30s (normal time in other runs) to 600s [!!!] after the following message appears:

=== Start of BnB marginalization at iteration 324 ==

This happens neither with the same box size and C1 symmetry nor 500px box size with I2 or C1 symmetry.

The job.log file yields the following information:

cryosparc2_compute/jobs/runcommon.py:738: RuntimeWarning: divide by zero encountered in float_scalars
run_old(*args, **kw)
cryosparc2_compute/jobs/runcommon.py:738: RuntimeWarning: divide by zero encountered in double_scalars
run_old(*args, **kw)
cryosparc2_compute/jobs/runcommon.py:738: RuntimeWarning: invalid value encountered in float_scalars
run_old(*args, **kw)
cryosparc2_compute/jobs/runcommon.py:738: RuntimeWarning: invalid value encountered in double_scalars
run_old(*args, **kw)
========= sending heartbeat

Did anybody else experience this? Are there known fixes?

Thanks in Advance,

Ben