summaryrefslogtreecommitdiff
path: root/training/conf/network
diff options
context:
space:
mode:
authorGustaf Rydholm <gustaf.rydholm@gmail.com>2023-09-02 01:53:55 +0200
committerGustaf Rydholm <gustaf.rydholm@gmail.com>2023-09-02 01:53:55 +0200
commitcf42f987a34d4de0db10f733fd532f949dd9c278 (patch)
tree1420cc62d2792a3921272f4a5aa39e40ef7f7ce9 /training/conf/network
parentabc2d60d69d115cdb34615d8bcb6c03ab6357141 (diff)
Update configs
Diffstat (limited to 'training/conf/network')
-rw-r--r--training/conf/network/conv_transformer.yaml59
-rw-r--r--training/conf/network/convnext.yaml17
2 files changed, 0 insertions, 76 deletions
diff --git a/training/conf/network/conv_transformer.yaml b/training/conf/network/conv_transformer.yaml
deleted file mode 100644
index 1e03946..0000000
--- a/training/conf/network/conv_transformer.yaml
+++ /dev/null
@@ -1,59 +0,0 @@
-_target_: text_recognizer.network.ConvTransformer
-encoder:
- _target_: text_recognizer.network.image_encoder.ImageEncoder
- encoder:
- _target_: text_recognizer.network.convnext.ConvNext
- dim: 16
- dim_mults: [2, 4, 8]
- depths: [3, 3, 6]
- downsampling_factors: [[2, 2], [2, 2], [2, 2]]
- pixel_embedding:
- _target_: "text_recognizer.network.transformer.embeddings.axial.\
- AxialPositionalEmbeddingImage"
- dim: &hidden_dim 128
- axial_shape: [7, 128]
- axial_dims: [64, 64]
-decoder:
- _target_: text_recognizer.network.text_decoder.TextDecoder
- hidden_dim: *hidden_dim
- num_classes: 58
- pad_index: 3
- decoder:
- _target_: text_recognizer.network.transformer.Decoder
- dim: *hidden_dim
- depth: 10
- block:
- _target_: text_recognizer.network.transformer.decoder_block.DecoderBlock
- self_attn:
- _target_: text_recognizer.network.transformer.Attention
- dim: *hidden_dim
- num_heads: 12
- dim_head: 64
- dropout_rate: &dropout_rate 0.2
- causal: true
- cross_attn:
- _target_: text_recognizer.network.transformer.Attention
- dim: *hidden_dim
- num_heads: 12
- dim_head: 64
- dropout_rate: *dropout_rate
- causal: false
- norm:
- _target_: text_recognizer.network.transformer.RMSNorm
- dim: *hidden_dim
- ff:
- _target_: text_recognizer.network.transformer.FeedForward
- dim: *hidden_dim
- dim_out: null
- expansion_factor: 2
- glu: true
- dropout_rate: *dropout_rate
- rotary_embedding:
- _target_: text_recognizer.network.transformer.RotaryEmbedding
- dim: 64
- token_pos_embedding:
- _target_: "text_recognizer.network.transformer.embeddings.fourier.\
- PositionalEncoding"
- dim: *hidden_dim
- dropout_rate: 0.1
- max_len: 89
diff --git a/training/conf/network/convnext.yaml b/training/conf/network/convnext.yaml
deleted file mode 100644
index 904bd56..0000000
--- a/training/conf/network/convnext.yaml
+++ /dev/null
@@ -1,17 +0,0 @@
-_target_: text_recognizer.network.convnext.ConvNext
-dim: 16
-dim_mults: [2, 4, 8]
-depths: [3, 3, 6]
-downsampling_factors: [[2, 2], [2, 2], [2, 2]]
-attn:
- _target_: text_recognizer.network.convnext.TransformerBlock
- attn:
- _target_: text_recognizer.network.convnext.Attention
- dim: 128
- heads: 4
- dim_head: 64
- scale: 8
- ff:
- _target_: text_recognizer.network.convnext.FeedForward
- dim: 128
- mult: 4