blob: dbd8a3b5adaafeae88f9b85bc4933cc6bdcf5428 (
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
79
80
81
|
# @package _global_
defaults:
- override /criterion: cross_entropy
- override /callbacks: htr
- override /datamodule: iam_lines
- override /network: vq_transformer
- override /model: lit_vq_transformer
- override /lr_scheduler: null
- override /optimizer: null
tags: [lines]
epochs: &epochs 200
ignore_index: &ignore_index 3
num_classes: &num_classes 57
max_output_len: &max_output_len 89
summary: [[1, 1, 56, 1024], [1, 89]]
logger:
wandb:
tags: ${tags}
# id: 342qvr1p
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.ReduceLROnPlateau
mode: min
factor: 0.8
patience: 10
threshold: 1.0e-4
threshold_mode: rel
cooldown: 0
min_lr: 1.0e-5
eps: 1.0e-8
verbose: false
interval: epoch
monitor: val/cer
datamodule:
batch_size: 8
train_fraction: 0.95
network:
input_dims: [1, 1, 56, 1024]
num_classes: *num_classes
pad_index: *ignore_index
encoder:
depth: 5
decoder:
depth: 6
pixel_embedding:
shape: [1, 127]
model:
max_output_len: *max_output_len
vq_loss_weight: 0.1
trainer:
gradient_clip_val: 1.0
max_epochs: *epochs
accumulate_grad_batches: 1
# resume_from_checkpoint: /home/aktersnurra/projects/text-recognizer/training/logs/runs/2022-06-27/00-37-40/checkpoints/last.ckpt
|
