Can we limit the depth of the MSA?

We have some users at our site folding large protein complexes using the multimer workflow, and have run into issues with memory exhausted on the GPUs even using the reduced databases. In the log file we notice it reporting what looks like a deep MSA:

`I0822 12:49:34.498847 140441650169664 pipeline.py:238] Final (deduplicated) MSA size: 11437 sequences.`

In the prediction stage, it doesn't look like the 11k sequences are used, as the msa dimensions are smaller:

`I0822 12:59:51.185315 140441650169664 model.py:166] Running predict with shape(feat) = {'aatype': (4205,), 'residue_index': (4205,), 'seq_length': (), 'msa': (3072, 4205), ...`

But this still looks larger than the MSA depth of 30 to 100 that I believe was identified as optimal in the original publication. Does this MSA depth appear abnormal, and if so is there anything we can do to restrict the MSA to a smaller number of sequences?