# Copyright 2016-2023 Cerebras Systems
# SPDX-License-Identifier: BSD-3-Clause
"""contains the Cerebras SGD implementation"""
from typing import Iterable
import torch
import cerebras.pytorch as cstorch
from .optimizer import Optimizer
[docs]class SGD(Optimizer):
"""
SGD optimizer implemented to conform to execution within the constraints
of the Cerebras WSE, including pre-initializing optimizer state
"""
def __init__(
self,
params: Iterable[torch.nn.Parameter],
lr: float,
momentum: float = 0,
dampening: float = 0,
weight_decay: float = 0,
nesterov: bool = False,
maximize: bool = False,
):
"""
Args:
params: Model parameters
lr: The learning rate to use
momentum: momentum factor
dampening: dampening for momentum
weight_decay: weight decay (L2 penalty)
nesterov: enables Nesterov momentum
"""
if lr < 0.0:
raise ValueError("Invalid learning rate: {}".format(lr))
if momentum < 0.0:
raise ValueError("Invalid momentum value: {}".format(momentum))
if weight_decay < 0.0:
raise ValueError(
"Invalid weight_decay value: {}".format(weight_decay)
)
if nesterov and (momentum <= 0 or dampening != 0):
raise ValueError(
f"Nesterov momentum requires a `momentum` and zero `dampening`. "
f"`momentum` was {momentum} and `dampening` was {dampening}."
)
defaults = dict(
lr=lr,
momentum=momentum,
dampening=dampening,
weight_decay=weight_decay,
nesterov=nesterov,
maximize=maximize,
)
super().__init__(params, defaults)
[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']:
if group['momentum'] != 0:
self.state[p]["momentum_buffer"] = 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:
loss = closure()
for group in self.param_groups:
lr = group["lr"]
weight_decay = group["weight_decay"]
momentum = group['momentum']
dampening = group["dampening"]
nesterov = group["nesterov"]
maximize = group["maximize"]
for p in group["params"]:
if p.grad is None:
continue
grad = p.grad
if grad.is_sparse:
raise RuntimeError("SGD does not support sparse gradients.")
grad = grad if not maximize else -grad
grad = grad + p * weight_decay
if momentum != 0:
buf = self.state[p]["momentum_buffer"]
buf.mul_(momentum).add_(grad, alpha=1.0 - dampening)
if nesterov:
grad.add_(buf, alpha=momentum)
else:
grad = buf
p.add_(-lr * grad)
return loss