Updates to configs

1 files changed, 20 insertions, 19 deletions

diff --git a/training/conf/experiment/conv_transformer_paragraphs_wp.yaml b/training/conf/experiment/conv_transformer_paragraphs_wp.yaml
index 499a609..ebaa17a 100644
--- a/training/conf/experiment/conv_transformer_paragraphs_wp.yaml
+++ b/training/conf/experiment/conv_transformer_paragraphs_wp.yaml
@@ -1,3 +1,5 @@
+# @package _global_
+
 defaults:
   - override /mapping: null
   - override /criterion: null
@@ -7,7 +9,6 @@ defaults:
   - override /lr_schedulers: null
   - override /optimizers: null
 
-
 epochs: &epochs 1000
 ignore_index: &ignore_index 1000
 num_classes: &num_classes 1006
@@ -17,17 +18,17 @@ summary: [[1, 1, 576, 640], [1, 451]]
 criterion:
   _target_: torch.nn.CrossEntropyLoss
   ignore_index: *ignore_index
-    
-mapping:
-  _target_: text_recognizer.data.word_piece_mapping.WordPieceMapping
-  num_features: 1000
-  tokens: iamdb_1kwp_tokens_1000.txt
-  lexicon: iamdb_1kwp_lex_1000.txt
-  data_dir: null
-  use_words: false
-  prepend_wordsep: false
-  special_tokens: [ <s>, <e>, <p> ]
-  extra_symbols: [ "\n" ]
+
+mapping: &mapping
+  mapping:
+    _target_: text_recognizer.data.mappings.word_piece_mapping.WordPieceMapping
+    num_features: 1000
+    tokens: iamdb_1kwp_tokens_1000.txt
+    lexicon: iamdb_1kwp_lex_1000.txt
+    use_words: false
+    prepend_wordsep: false
+    special_tokens: [ <s>, <e>, <p> ]
+    extra_symbols: [ "\n" ]
 
 callbacks:
   stochastic_weight_averaging:
@@ -41,7 +42,7 @@ callbacks:
 optimizers:
   madgrad:
     _target_: madgrad.MADGRAD
-    lr: 3.0e-4
+    lr: 1.0e-4
     momentum: 0.9
     weight_decay: 0
     eps: 1.0e-6
@@ -71,13 +72,13 @@ lr_schedulers:
 
 datamodule:
   _target_: text_recognizer.data.iam_extended_paragraphs.IAMExtendedParagraphs
-  batch_size: 4
   num_workers: 12
-  train_fraction: 0.8
-  augment: true
+  train_fraction: 0.9
   pin_memory: true
-  word_pieces: true
-  resize: null
+  transform: transform/paragraphs.yaml
+  test_transform: transform/paragraphs.yaml
+  target_transform: target_transform/word_piece.yaml
+  << : *mapping
 
 network:
   _target_: text_recognizer.networks.conv_transformer.ConvTransformer
@@ -138,7 +139,7 @@ trainer:
   stochastic_weight_avg: true
   auto_scale_batch_size: binsearch
   auto_lr_find: false
-  gradient_clip_val: 0.0
+  gradient_clip_val: 0.5
   fast_dev_run: false
   gpus: 1
   precision: 16