Functions — github.com/KellerJordan/Muon

CodeHub MCPcopy Create free account

hub / github.com/KellerJordan/Muon / functions

Functions11 in github.com/KellerJordan/Muon

⨍ Functions 11 ◇ Types & classes 4

↓ 4 callersFunctionmuon_update

(grad, momentum, beta=0.95, ns_steps=5, nesterov=True)

↓ 3 callersMethod__init__

(self, params, lr=0.02, weight_decay=0, momentum=0.95)

↓ 2 callersFunctionadam_update

(grad, buf1, buf2, step, betas, eps)

↓ 1 callersFunctionzeropower_via_newtonschulz5

Newton-Schulz iteration to compute the zeroth power / orthogonalization of G. We opt to use a quintic iteration whose coefficients are select

(self, params, lr=0.02, weight_decay=0, momentum=0.95)

(self, param_groups)

(self, param_groups)

(self, closure=None)

(self, closure=None)

(self, closure=None)

(self, closure=None)