diff --git a/combo/config.template.json b/combo/config.template.json
new file mode 100644
index 0000000000000000000000000000000000000000..7baf6547e1e49c0d0f3d0e441546cd7893329506
--- /dev/null
+++ b/combo/config.template.json
@@ -0,0 +1,307 @@
+{
+ "model": {
+ "type": "semantic_multitask",
+ "parameters": {
+ "dependency_relation": {
+ "type": "combo_dependency_parsing_from_vocab",
+ "parameters": {
+ "dependency_projection_layer": {
+ "type": "linear_layer",
+ "parameters": {
+ "activation": { "type": "tanh", "parameters": {} },
+ "dropout_rate": 0.25,
+ "in_features": 1024,
+ "out_features": 128
+ }
+ },
+ "head_predictor": {
+ "type": "head_prediction",
+ "parameters": {
+ "cycle_loss_n": 0,
+ "dependency_projection_layer": {
+ "type": "linear_layer",
+ "parameters": {
+ "activation": { "type": "tanh", "parameters": {} },
+ "in_features": 1024,
+ "out_features": 512
+ }
+ },
+ "head_projection_layer": {
+ "type": "linear_layer",
+ "parameters": {
+ "activation": { "type": "tanh", "parameters": {} },
+ "in_features": 1024,
+ "out_features": 512
+ }
+ }
+ }
+ },
+ "head_projection_layer": {
+ "type": "linear_layer",
+ "parameters": {
+ "activation": { "type": "tanh", "parameters": {} },
+ "dropout_rate": 0.25,
+ "in_features": 1024,
+ "out_features": 128
+ }
+ },
+ "vocab_namespace": "deprel_labels"
+ }
+ },
+ "lemmatizer": {
+ "type": "combo_lemma_predictor_from_vocab",
+ "parameters": {
+ "activations": [
+ { "type": "gelu", "parameters": {} },
+ { "type": "gelu", "parameters": {} },
+ { "type": "gelu", "parameters": {} },
+ { "type": "linear", "parameters": {} }
+ ],
+ "char_vocab_namespace": "token_characters",
+ "dilation": [1, 2, 4, 1],
+ "embedding_dim": 300,
+ "filters": [256, 256, 256],
+ "input_projection_layer": {
+ "type": "linear_layer",
+ "parameters": {
+ "activation": { "type": "tanh", "parameters": {} },
+ "dropout_rate": 0.25,
+ "in_features": 1024,
+ "out_features": 32
+ }
+ },
+ "kernel_size": [3, 3, 3, 1],
+ "lemma_vocab_namespace": "lemma_characters",
+ "padding": [1, 2, 4, 0],
+ "stride": [1, 1, 1, 1]
+ }
+ },
+ "loss_weights": {
+ "deprel": 0.8,
+ "feats": 0.2,
+ "head": 0.2,
+ "lemma": 0.05,
+ "semrel": 0.05,
+ "upostag": 0.05,
+ "xpostag": 0.05
+ },
+ "morphological_feat": {
+ "type": "combo_morpho_from_vocab",
+ "parameters": {
+ "activations": [
+ { "type": "tanh", "parameters": {} },
+ { "type": "linear", "parameters": {} }
+ ],
+ "dropout": [0.25, 0.0],
+ "hidden_dims": [128],
+ "input_dim": 1024,
+ "num_layers": 2,
+ "vocab_namespace": "feats_labels"
+ }
+ },
+ "regularizer": {
+ "type": "base_regularizer",
+ "parameters": {
+ "regexes": [
+ [
+ ".*conv1d.*",
+ { "type": "l2_regularizer", "parameters": { "alpha": 1e-6 } }
+ ],
+ [
+ ".*forward.*",
+ { "type": "l2_regularizer", "parameters": { "alpha": 1e-6 } }
+ ],
+ [
+ ".*backward.*",
+ { "type": "l2_regularizer", "parameters": { "alpha": 1e-6 } }
+ ],
+ [
+ ".*char_embed.*",
+ { "type": "l2_regularizer", "parameters": { "alpha": 1e-5 } }
+ ]
+ ]
+ }
+ },
+ "seq_encoder": {
+ "type": "combo_encoder",
+ "parameters": {
+ "layer_dropout_probability": 0.33,
+ "stacked_bilstm": {
+ "type": "combo_stacked_bilstm",
+ "parameters": {
+ "hidden_size": 512,
+ "input_size": 164,
+ "layer_dropout_probability": 0.33,
+ "num_layers": 2,
+ "recurrent_dropout_probability": 0.33
+ }
+ }
+ }
+ },
+ "text_field_embedder": {
+ "type": "base_text_field_embedder",
+ "parameters": {
+ "token_embedders": {
+ "char": {
+ "type": "char_embeddings_token_embedder",
+ "parameters": {
+ "dilated_cnn_encoder": {
+ "type": "dilated_cnn",
+ "parameters": {
+ "activations": [
+ { "type": "gelu", "parameters": {} },
+ { "type": "gelu", "parameters": {} },
+ { "type": "linear", "parameters": {} }
+ ],
+ "dilation": [1, 2, 4],
+ "filters": [512, 256, 64],
+ "input_dim": 64,
+ "kernel_size": [3, 3, 3],
+ "padding": [1, 2, 4],
+ "stride": [1, 1, 1]
+ }
+ },
+ "embedding_dim": 64
+ }
+ },
+ "token": {
+ "type": "transformers_word_embedder",
+ "parameters": { "projection_dim": 100 }
+ }
+ }
+ }
+ },
+ "upos_tagger": {
+ "type": "feedforward_predictor_from_vocab",
+ "parameters": {
+ "vocab_namespace": "upostag_labels",
+ "input_dim": 1024,
+ "num_layers": 2,
+ "hidden_dims": [64],
+ "activations": [
+ { "type": "tanh", "parameters": {} },
+ { "type": "linear", "parameters": {} }
+ ],
+ "dropout": [0.25, 0.0]
+ }
+ },
+ "xpos_tagger": {
+ "type": "feedforward_predictor_from_vocab",
+ "parameters": {
+ "vocab_namespace": "xpostag_labels",
+ "input_dim": 1024,
+ "num_layers": 2,
+ "hidden_dims": [64],
+ "activations": [
+ { "type": "tanh", "parameters": {} },
+ { "type": "linear", "parameters": {} }
+ ],
+ "dropout": [0.25, 0.0]
+ }
+ }
+ }
+ },
+ "data_loader": {
+ "type": "simple_data_loader_from_dataset_reader",
+ "parameters": {
+ "reader": {
+ "type": "conllu_dataset_reader",
+ "parameters": {
+ "features": ["token", "char"],
+ "tokenizer": {
+ "type": "lambo_tokenizer"
+ },
+ "lemma_indexers": {
+ "char": {
+ "type": "characters_const_padding_token_indexer",
+ "parameters": {
+ "tokenizer": {
+ "type": "character_tokenizer",
+ "parameters": {
+ "end_tokens": ["__END__"],
+ "start_tokens": ["__START__"]
+ }
+ },
+ "min_padding_length": 32,
+ "namespace": "lemma_characters"
+ }
+ }
+ },
+ "targets": ["deprel", "feats", "head", "lemma", "upostag", "xpostag"],
+ "token_indexers": {
+ "char": {
+ "type": "characters_const_padding_token_indexer",
+ "parameters": {
+ "tokenizer": {
+ "type": "character_tokenizer",
+ "parameters": {
+ "end_tokens": ["__END__"],
+ "start_tokens": ["__START__"]
+ }
+ },
+ "min_padding_length": 32
+ }
+ },
+ "token": {
+ "type": "pretrained_transformer_mismatched_fixed_token_indexer",
+ "parameters": { "model_name": "allegro/herbert-base-cased" }
+ }
+ },
+ "use_sem": false
+ }
+ },
+ "batch_size": 1,
+ "shuffle": true,
+ "batches_per_epoch": 64,
+ "quiet": false
+ }
+ },
+ "dataset_reader": {
+ "type": "conllu_dataset_reader",
+ "parameters": {
+ "features": ["token", "char"],
+ "tokenizer": {
+ "type": "lambo_tokenizer"
+ },
+ "lemma_indexers": {
+ "char": {
+ "type": "characters_const_padding_token_indexer",
+ "parameters": {
+ "tokenizer": {
+ "type": "character_tokenizer",
+ "parameters": {
+ "end_tokens": ["__END__"],
+ "start_tokens": ["__START__"]
+ }
+ },
+ "min_padding_length": 32,
+ "namespace": "lemma_characters"
+ }
+ }
+ },
+ "targets": ["deprel", "feats", "head", "lemma", "upostag", "xpostag"],
+ "token_indexers": {
+ "char": {
+ "type": "characters_const_padding_token_indexer",
+ "parameters": {
+ "tokenizer": {
+ "type": "character_tokenizer",
+ "parameters": {
+ "end_tokens": ["__END__"],
+ "start_tokens": ["__START__"]
+ }
+ },
+ "min_padding_length": 32
+ }
+ },
+ "token": {
+ "type": "pretrained_transformer_mismatched_fixed_token_indexer",
+ "parameters": { "model_name": "allegro/herbert-base-cased" }
+ }
+ },
+ "use_sem": false
+ }
+ },
+ "training": {},
+ "model_name": "allegro/herbert-base-cased"
+}
diff --git a/combo/modules/archival.py b/combo/modules/archival.py
index 496753f711f60fce9555d724fe72c7e49c7a8435..63a4cd831aa9f27ce6f4c31f6cee6f7f670552c8 100644
--- a/combo/modules/archival.py
+++ b/combo/modules/archival.py
@@ -81,7 +81,7 @@ def archive(model: Model,
with (TemporaryDirectory(os.path.join('tmp')) as t,
BytesIO() as out_stream,
tarfile.open(os.path.join(serialization_dir, 'model.tar.gz'), 'w|gz') as tar_file):
- add_to_tar(tar_file, out_stream, json.dumps(parameters).encode(), 'config.json')
+ add_to_tar(tar_file, out_stream, json.dumps(parameters).encode(), 'config.template.json')
weights_path = os.path.join(t, 'weights.th')
torch.save(model.state_dict(), weights_path)
tar_file.add(weights_path, 'weights.th')
@@ -100,7 +100,7 @@ def extracted_archive(resolved_archive_file, cleanup=True):
with tarfile.open(resolved_archive_file) as archive:
subdir_and_files = [
tarinfo for tarinfo in archive.getmembers()
- if (any([tarinfo.name.endswith(f) for f in ['config.json', 'weights.th']])
+ if (any([tarinfo.name.endswith(f) for f in ['config.template.json', 'weights.th']])
or 'vocabulary' in tarinfo.name)
]
for f in subdir_and_files:
@@ -127,7 +127,7 @@ def load_archive(url_or_filename: Union[PathLike, str],
with extracted_archive(rarchive_file) as archive_file:
model = Model.load(archive_file, cuda_device=cuda_device)
- config_path = os.path.join(archive_file, 'config.json')
+ config_path = os.path.join(archive_file, 'config.template.json')
with open(config_path, 'r') as f:
config = json.load(f)