default shared memory too low

[robert-s-lee]

workaround is to set training.num_workers=2

tried below and shm_size=512 causes the work to fail without any error messages.

        self.my_work = LitBashWork(
          cloud_compute=L.CloudCompute("gpu", shm_size=512), 
          cloud_build_config=FiftyOneBuildConfig(),
          )