blob: d3311fd75047e4cf33726da3cf8a6732b17adf52 (
plain)
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
|
# @package _global_
defaults:
- override /criterion: cross_entropy
- override /callbacks: htr
- override /datamodule: iam_extended_paragraphs
- override /network: conv_transformer
- override /model: lit_transformer
- override /lr_scheduler: null
- override /optimizer: null
epochs: &epochs 600
num_classes: &num_classes 58
ignore_index: &ignore_index 3
max_output_len: &max_output_len 682
summary: [[1, 1, 576, 640], [1, 682]]
criterion:
ignore_index: *ignore_index
label_smoothing: 0.05
callbacks:
stochastic_weight_averaging:
_target_: pytorch_lightning.callbacks.StochasticWeightAveraging
swa_epoch_start: 0.75
swa_lrs: 1.0e-5
annealing_epochs: 10
annealing_strategy: cos
device: null
optimizer:
_target_: torch.optim.RAdam
lr: 3.0e-4
betas: [0.9, 0.999]
weight_decay: 0
eps: 1.0e-8
lr_scheduler:
_target_: torch.optim.lr_scheduler.OneCycleLR
max_lr: 3.0e-4
total_steps: null
epochs: *epochs
steps_per_epoch: 3358
pct_start: 0.15
anneal_strategy: cos
cycle_momentum: true
base_momentum: 0.85
max_momentum: 0.95
div_factor: 25.0
final_div_factor: 10000.0
three_phase: true
last_epoch: -1
verbose: false
interval: step
monitor: val/cer
datamodule:
batch_size: 6
train_fraction: 0.95
network:
input_dims: [1, 1, 576, 640]
num_classes: *num_classes
pad_index: *ignore_index
encoder:
depth: 6
decoder:
depth: 5
pixel_embedding:
shape: [18, 80]
model:
max_output_len: *max_output_len
trainer:
gradient_clip_val: 0.5
max_epochs: *epochs
accumulate_grad_batches: 1
|
