I often get the same error message when doing uccsd(t) calculations. I find that often just using a different starting guess for the hartree-fock calculation will cause the problem to go away (even though the hartree-fock calculation appears to converge to the same solution in both cases).