Skip to content

Instantly share code, notes, and snippets.

@kastnerkyle
Forked from crowsonkb/adamw_finetune.py
Created August 1, 2022 16:41
Show Gist options
  • Save kastnerkyle/82e3da422a649461307fc21110af6e68 to your computer and use it in GitHub Desktop.
Save kastnerkyle/82e3da422a649461307fc21110af6e68 to your computer and use it in GitHub Desktop.
import math
import torch
from torch import optim
class AdamWFinetune(optim.Optimizer):
r"""Implements AdamW algorithm with optional weight decay toward the starting value, to
prevent overfitting to the new dataset during fine-tuning.
The original Adam algorithm was proposed in `Adam: A Method for Stochastic Optimization`_.
The AdamW variant was proposed in `Decoupled Weight Decay Regularization`_. Unlike that
variant, this optimizer follows the PyTorch behavior of scaling the weight decay
coefficients by the learning rate.
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-8)
weight_decay (float, optional): weight decay coefficient (default: 0)
weight_decay_toward_start (float, optional): weight decay toward starting
value coefficient (default: 0)
amsgrad (boolean, optional): whether to use the AMSGrad variant of this
algorithm from the paper `On the Convergence of Adam and Beyond`_
(default: False)
maximize (bool, optional): maximize the params based on the objective, instead of
minimizing (default: False)
.. _Adam\: A Method for Stochastic Optimization:
https://arxiv.org/abs/1412.6980
.. _Decoupled Weight Decay Regularization:
https://arxiv.org/abs/1711.05101
.. _On the Convergence of Adam and Beyond:
https://openreview.net/forum?id=ryQu7f-RZ
"""
def __init__(self, params, lr=1e-3, betas=(0.9, 0.999), eps=1e-8,
weight_decay=0., weight_decay_toward_start=0., amsgrad=False, *,
maximize=False):
if not 0.0 <= lr:
raise ValueError("Invalid learning rate: {}".format(lr))
if not 0.0 <= eps:
raise ValueError("Invalid epsilon value: {}".format(eps))
if not 0.0 <= betas[0] < 1.0:
raise ValueError("Invalid beta parameter at index 0: {}".format(betas[0]))
if not 0.0 <= betas[1] < 1.0:
raise ValueError("Invalid beta parameter at index 1: {}".format(betas[1]))
if not 0.0 <= weight_decay:
raise ValueError("Invalid weight_decay value: {}".format(weight_decay))
if not 0.0 <= weight_decay_toward_start:
raise ValueError("Invalid weight_decay_toward_start value: {}".format(weight_decay_toward_start))
defaults = dict(lr=lr, betas=betas, eps=eps,
weight_decay=weight_decay,
weight_decay_toward_start=weight_decay_toward_start,
amsgrad=amsgrad, maximize=maximize)
super().__init__(params, defaults)
def __setstate__(self, state):
super().__setstate__(state)
for group in self.param_groups:
group.setdefault('amsgrad', False)
group.setdefault('maximize', False)
@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:
for p in group['params']:
if p.grad is None:
continue
# Perform optimization step
grad = p.grad
if grad.is_sparse:
raise RuntimeError('AdamW does not support sparse gradients')
amsgrad = group['amsgrad']
state = self.state[p]
# State initialization
if len(state) == 0:
state['step'] = 0
# Starting value
state['start'] = p.clone()
# Exponential moving average of gradient values
state['exp_avg'] = torch.zeros_like(p, memory_format=torch.preserve_format)
# Exponential moving average of squared gradient values
state['exp_avg_sq'] = torch.zeros_like(p, memory_format=torch.preserve_format)
if amsgrad:
# Maintains max of all exp. moving avg. of sq. grad. values
state['max_exp_avg_sq'] = torch.zeros_like(p, memory_format=torch.preserve_format)
# Decay toward 0
p.mul_(1 - group['lr'] * group['weight_decay'])
# Decay toward starting value
p.lerp_(state['start'], group['lr'] * group['weight_decay_toward_start'])
exp_avg, exp_avg_sq = state['exp_avg'], state['exp_avg_sq']
if amsgrad:
max_exp_avg_sq = state['max_exp_avg_sq']
beta1, beta2 = group['betas']
state['step'] += 1
bias_correction1 = 1 - beta1 ** state['step']
bias_correction2 = 1 - beta2 ** state['step']
# Decay the first and second moment running average coefficient
exp_avg.mul_(beta1).add_(grad, alpha=1 - beta1)
exp_avg_sq.mul_(beta2).addcmul_(grad, grad, value=1 - beta2)
if amsgrad:
# Maintains the maximum of all 2nd moment running avg. till now
torch.maximum(max_exp_avg_sq, exp_avg_sq, out=max_exp_avg_sq)
# Use the max. for normalizing running avg. of gradient
denom = (max_exp_avg_sq.sqrt() / math.sqrt(bias_correction2)).add_(group['eps'])
else:
denom = (exp_avg_sq.sqrt() / math.sqrt(bias_correction2)).add_(group['eps'])
step_size = group['lr'] / bias_correction1
p.addcdiv_(exp_avg, denom, value=step_size if group['maximize'] else -step_size)
return loss
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment