File size: 4,572 Bytes
3cd1076 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 | #!/usr/bin/env python
# -*- coding: utf-8 -*-
import math
import torch
from torch.optim.optimizer import Optimizer
class Lamb(Optimizer):
r"""Implements Lamb algorithm.
It has been proposed in `Large Batch Optimization for Deep Learning:
Training BERT in 76 minutes`
https://arxiv.org/abs/1904.00962
Note:
Reference code: https://github.com/cybertronai/pytorch-lamb
"""
def __init__(
self,
params,
lr: float = 1e-3,
betas = (0.9, 0.999),
eps: float = 1e-6,
weight_decay: float = 0,
clamp_value: float = 10,
adam: bool = False,
debias: bool = False,
) -> None:
if lr <= 0.0:
raise ValueError("Invalid learning rate: {}".format(lr))
if eps < 0.0:
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 weight_decay < 0:
raise ValueError(
"Invalid weight_decay value: {}".format(weight_decay)
)
if clamp_value < 0.0:
raise ValueError("Invalid clamp value: {}".format(clamp_value))
defaults = dict(lr=lr, betas=betas, eps=eps, weight_decay=weight_decay)
self.clamp_value = clamp_value
self.adam = adam
self.debias = debias
super(Lamb, self).__init__(params, defaults)
def step(self, closure = None):
r"""Performs a single optimization step.
Arguments:
closure: 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:
for p in group["params"]:
if p.grad is None:
continue
grad = p.grad.data
if grad.is_sparse:
msg = (
"Lamb does not support sparse gradients, "
"please consider SparseAdam instead"
)
raise RuntimeError(msg)
state = self.state[p]
# State initialization
if len(state) == 0:
state["step"] = 0
# 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
)
exp_avg, exp_avg_sq = state["exp_avg"], state["exp_avg_sq"]
beta1, beta2 = group["betas"]
state["step"] += 1
# Decay the first and second moment running average coefficient
# m_t
exp_avg.mul_(beta1).add_(grad, alpha=1 - beta1)
# v_t
exp_avg_sq.mul_(beta2).addcmul_(grad, grad, value=1 - beta2)
# Paper v3 does not use debiasing.
if self.debias:
bias_correction = math.sqrt(1 - beta2 ** state["step"])
bias_correction /= 1 - beta1 ** state["step"]
else:
bias_correction = 1
# Apply bias to lr to avoid broadcast.
step_size = group["lr"] * bias_correction
weight_norm = torch.norm(p.data).clamp(0, self.clamp_value)
adam_step = exp_avg / exp_avg_sq.sqrt().add(group["eps"])
if group["weight_decay"] != 0:
adam_step.add_(p.data, alpha=group["weight_decay"])
adam_norm = torch.norm(adam_step)
if weight_norm == 0 or adam_norm == 0:
trust_ratio = 1
else:
trust_ratio = weight_norm / adam_norm
state["weight_norm"] = weight_norm
state["adam_norm"] = adam_norm
state["trust_ratio"] = trust_ratio
if self.adam:
trust_ratio = 1
p.data.add_(adam_step, alpha=-step_size * trust_ratio)
return loss |