#
# Cerebras implementation of RAdam optimizer. Adapted from the `torch.optim.RAdam` implementation.
#
# Copyright 2016-2023 Cerebras Systems
# SPDX-License-Identifier: BSD-3-Clause
#
from typing import Iterable, Tuple
import torch
import cerebras_pytorch as cstorch
from .optimizer import Optimizer
[docs]class RAdam(Optimizer):
r"""RAdam optimizer implemented to conform to execution within the
constraints of the Cerebras WSE.
Args:
params (iterable): iterable of parameters to optimize or dicts defining
parameter groups
lr (float, optional): learning rate (default: 1e-3)
betas (Tuple[float, float], optional): coefficients used for computing
running averages of gradient and its square (default: (0.9, 0.999))
eps (float, optional): term added to the denominator to improve
numerical stability (default: 1e-6)
weight_decay (float, optional): weight decay (L2 penalty) (default: 0)
"""
[docs] def __init__(
self,
params: Iterable[torch.nn.Parameter],
lr: float = 1e-3,
betas: Tuple[float, float] = (0.9, 0.999),
eps: float = 1e-6,
weight_decay: float = 0.0,
):
if lr < 0.0:
raise ValueError(f"Invalid learning rate: {lr} - should be >= 0.0")
if not 0.0 <= betas[0] < 1.0:
raise ValueError(
f"Invalid beta parameter: {betas[0]} - should be in [0.0, 1.0]"
)
if not 0.0 <= betas[1] < 1.0:
raise ValueError(
f"Invalid beta parameter: {betas[1]} - should be in [0.0, 1.0]"
)
if weight_decay < 0.0:
raise ValueError(
f"Invalid weight_decay value: {weight_decay} - should be >= 0.0"
)
if eps < 0.0:
raise ValueError(f"Invalid epsilon value: {eps} - should be >= 0.0")
defaults = dict(lr=lr, betas=betas, eps=eps, weight_decay=weight_decay,)
super().__init__(params, defaults, enable_global_step=True)
[docs] def state_names_to_sparsify(self):
return ["exp_avg", "exp_avg_sq"]
[docs] def preinitialize(self):
"""
Allocates tensors for the optimizer state to allow direct compilation
of the model before the first step.
"""
for group in self.param_groups:
for p in group["params"]:
state = self.state[p]
# State initialization
# Exponential moving average of gradient values
state["exp_avg"] = cstorch.zeros_like(p)
# Exponential moving average of squared gradient values
state["exp_avg_sq"] = cstorch.zeros_like(p)
@torch.no_grad()
def step(self, closure=None):
"""Performs a single optimization step.
Args:
closure (callable, optional): A closure that reevaluates the model
and returns the loss.
"""
loss = None
if closure is not None:
with torch.enable_grad():
loss = closure()
for group in self.param_groups:
beta1, beta2 = group['betas']
if not isinstance(beta1, torch.Tensor):
beta1 = torch.tensor(beta1)
if not isinstance(beta2, torch.Tensor):
beta2 = torch.tensor(beta2)
weight_decay = group["weight_decay"]
eps = group["eps"]
for p in group['params']:
if p.grad is not None:
grad = p.grad
state = self.state[p]
exp_avg, exp_avg_sq = state["exp_avg"], state["exp_avg_sq"]
global_step = self.increment_global_step(p)
if weight_decay > 0.0:
grad = grad.add(p, alpha=weight_decay)
beta1t = torch.pow(beta1.to(p.device), global_step)
beta2t = torch.pow(beta2.to(p.device), global_step)
bias_correction1 = 1 - beta1t
bias_correction2 = 1 - beta2t
# Decay the first and second moment running average coefficient
# In-place operations to update the averages at the same time.
exp_avg.mul_(beta1).add_(grad, alpha=1.0 - beta1)
exp_avg_sq.mul_(beta2).addcmul_(
grad, grad, value=1.0 - beta2
)
# correcting bias for the first moving moment
update = exp_avg / bias_correction1
# maximum length of the approximated SMA
rho_inf = 2 / (1 - beta2) - 1
# compute the length of the approximated SMA
rho_t = (
rho_inf - 2 * global_step * beta2t / bias_correction2
)
one = torch.tensor(1.0).to(p.device)
five = torch.tensor(5.0).to(p.device)
# Compute the variance rectification term and update parameters accordingly
rect = torch.where(
torch.gt(rho_t, five),
torch.sqrt(
(rho_t - 4.0)
* (rho_t - 2.0)
* rho_inf
/ ((rho_inf - 4.0) * (rho_inf - 2.0) * rho_t)
),
one,
)
adaptive_lr = torch.where(
torch.gt(rho_t, five),
torch.sqrt(bias_correction2)
/ exp_avg_sq.sqrt().add_(eps),
one,
)
update *= rect
update *= adaptive_lr
update *= group["lr"]
p.sub_(update)
return loss