深度学习lr scheduler 介绍_lr_scheduler

lr_scheduler在深度学习模型中经常遇到，虽粗通其理，然未解其中奥秘。

简单整理，冀假以时日，略加参悟。

1.WarmupLinearScheduler

代码参考自https://github.com/huggingface/transformers/blob/main/src/transformers/optimization.py
的get_linear_schedule_with_warmup部分。

import matplotlib.pyplot as plt
t_total = 10000
warmup_steps = 1000
lr = 1e-4

def lr_lambda(step):
    if step < warmup_steps:
        return float(step) / float(max(1, warmup_steps))
    return max(0.0, float(t_total - step) / float(
        max(1.0, t_total - warmup_steps)))
        
lrs = []
for step in range(t_total):
    lrs.append(lr*lr_lambda(step))
plt.plot(lrs)
plt.show()
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16

参考文献

[1]https://github.com/huggingface/transformers/blob/main/src/transformers/optimization.py
[2] pytorch how-to-adjust-learning-rate