AssertionError: If using tupled rtol it must have the same length as the tuple y0

[JiafengtTang]

when I run the training, the "AssertionError: If using tupled rtol it must have the same length as the tuple y0" occurs.

[binvec]

change line 58-59 in lib/layers/cnf.py:

atol=[self.atol, self.atol] + [1e20] * len(reg_states) if self.solver in ['dopri5', 'bosh3'] else self.atol,
rtol=[self.rtol, self.rtol] + [1e20] * len(reg_states) if self.solver in ['dopri5', 'bosh3'] else self.rtol,

to

atol= self.atol + 1e20 * len(reg_states) if self.solver in ['dopri5', 'bosh3'] else self.atol,
rtol= self.rtol + 1e20 * len(reg_states) if self.solver in ['dopri5', 'bosh3'] else self.rtol,

[bfs18]

change line 58-59 in lib/layers/cnf.py:

atol=[self.atol, self.atol] + [1e20] * len(reg_states) if self.solver in ['dopri5', 'bosh3'] else self.atol,
rtol=[self.rtol, self.rtol] + [1e20] * len(reg_states) if self.solver in ['dopri5', 'bosh3'] else self.rtol,

to

atol= self.atol + 1e20 * len(reg_states) if self.solver in ['dopri5', 'bosh3'] else self.atol,
rtol= self.rtol + 1e20 * len(reg_states) if self.solver in ['dopri5', 'bosh3'] else self.rtol,

This is incorrect.

The following code is correct.

            state_t = odeint(
                self.odefunc,
                (z, _logpz) + reg_states,
                integration_times.to(z),
                atol=[self.atol] * 2 + [1e20] * len(reg_states) if self.solver in ['dopri5', 'bosh3'] else self.atol,
                rtol=[self.rtol] * 2 + [1e20] * len(reg_states) if self.solver in ['dopri5', 'bosh3'] else self.rtol,
                adjoint_atol=self.atol,
                adjoint_rtol=self.rtol,
                method=self.solver,
                options=self.solver_options,
                adjoint_options=self.adjoint_options
            )