summaryrefslogtreecommitdiff
path: root/training/conf
diff options
context:
space:
mode:
Diffstat (limited to 'training/conf')
-rw-r--r--training/conf/experiment/conv_transformer_paragraphs.yaml20
1 files changed, 10 insertions, 10 deletions
diff --git a/training/conf/experiment/conv_transformer_paragraphs.yaml b/training/conf/experiment/conv_transformer_paragraphs.yaml
index 859117f..7a72a1a 100644
--- a/training/conf/experiment/conv_transformer_paragraphs.yaml
+++ b/training/conf/experiment/conv_transformer_paragraphs.yaml
@@ -89,7 +89,7 @@ network:
stochastic_dropout_rate: 0.2
bn_momentum: 0.99
bn_eps: 1.0e-3
- depth: 7
+ depth: 5
decoder:
depth: 6
_target_: text_recognizer.networks.transformer.layers.Decoder
@@ -116,15 +116,15 @@ network:
pixel_pos_embedding:
_target_: text_recognizer.networks.transformer.embeddings.axial.AxialPositionalEmbedding
dim: *hidden_dim
- shape: &shape [18, 20]
- axial_encoder:
- _target_: text_recognizer.networks.transformer.axial_attention.encoder.AxialEncoder
- dim: *hidden_dim
- heads: 4
- shape: *shape
- depth: 2
- dim_head: 64
- dim_index: 1
+ shape: &shape [36, 40]
+ axial_encoder: null
+ # _target_: text_recognizer.networks.transformer.axial_attention.encoder.AxialEncoder
+ # dim: *hidden_dim
+ # heads: 4
+ # shape: *shape
+ # depth: 2
+ # dim_head: 64
+ # dim_index: 1
model:
_target_: text_recognizer.models.transformer.TransformerLitModel