torch.optim.functional.muon

torch.optim.functional.muon(params, grads, muon_momentum_bufs, *, foreach=None, lr, weight_decay, momentum, nesterov, ns_coefficients, ns_steps, eps, adjust_lr_fn, has_complex)

Functional API that performs Muon algorithm computation.

This function updates the provided parameters and optimizer state in place. The caller must initialize and retain optimizer state. Unless intentionally constructing a differentiable update with a supported differentiable=True argument, call this function under torch.no_grad. See Functional optimizer API for the common functional optimizer contract and examples, and Muon for algorithm details.