blob: b34dd117f4ca8e39b027e204560db79f3a044f4f (
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
|
# @package _global_
defaults:
- override /mapping: null
- override /network: null
- override /model: null
mapping:
_target_: text_recognizer.data.emnist_mapping.EmnistMapping
extra_symbols: [ "\n" ]
datamodule:
word_pieces: false
batch_size: 8
criterion:
ignore_index: 3
network:
_target_: text_recognizer.networks.vq_transformer.VqTransformer
input_dims: [1, 576, 640]
encoder_dim: 64
hidden_dim: 64
dropout_rate: 0.1
num_classes: 58
pad_index: 3
no_grad: false
encoder:
_target_: text_recognizer.networks.vqvae.vqvae.VQVAE
hidden_dim: 128
embedding_dim: 64
num_embeddings: 1024
decay: 0.99
encoder:
_target_: text_recognizer.networks.vqvae.encoder.Encoder
in_channels: 1
hidden_dim: 64
channels_multipliers: [1, 1, 2, 2]
dropout_rate: 0.0
decoder:
_target_: text_recognizer.networks.vqvae.decoder.Decoder
out_channels: 1
hidden_dim: 64
channels_multipliers: [2, 2, 1, 1]
dropout_rate: 0.0
decoder:
_target_: text_recognizer.networks.transformer.Decoder
dim: 64
depth: 2
num_heads: 4
attn_fn: text_recognizer.networks.transformer.attention.Attention
attn_kwargs:
dim_head: 32
dropout_rate: 0.2
norm_fn: torch.nn.LayerNorm
ff_fn: text_recognizer.networks.transformer.mlp.FeedForward
ff_kwargs:
dim_out: null
expansion_factor: 4
glu: true
dropout_rate: 0.2
cross_attend: true
pre_norm: true
rotary_emb: null
# pretrained_encoder_path: "training/logs/runs/2021-09-13/08-35-57/checkpoints/epoch=98.ckpt"
model:
_target_: text_recognizer.models.vq_transformer.VqTransformerLitModel
start_token: <s>
end_token: <e>
pad_token: <p>
max_output_len: 682
# max_output_len: 451
|
