refactor: torch-based components (#1617)

deeppavlov · Jun 28, 2023 · 59b7557 · 59b7557
1 parent 1d216db
commit 59b7557
Show file tree

Hide file tree

Showing 29 changed files with 301 additions and 1,418 deletions.
diff --git a/deeppavlov/configs/classifiers/rusentiment_bert.json b/deeppavlov/configs/classifiers/rusentiment_bert.json
@@ -63,7 +63,6 @@
         "class_name": "torch_transformers_classifier",
         "n_classes": "#classes_vocab.len",
         "return_probas": true,
-        "one_hot_labels": true,
         "pretrained_bert": "{TRANSFORMER}",
         "save_path": "{MODEL_PATH}/model",
         "load_path": "{MODEL_PATH}/model",

diff --git a/deeppavlov/configs/classifiers/rusentiment_convers_bert.json b/deeppavlov/configs/classifiers/rusentiment_convers_bert.json
@@ -63,7 +63,6 @@
         "class_name": "torch_transformers_classifier",
         "n_classes": "#classes_vocab.len",
         "return_probas": true,
-        "one_hot_labels": true,
         "pretrained_bert": "{TRANSFORMER}",
         "save_path": "{MODEL_PATH}/model",
         "load_path": "{MODEL_PATH}/model",

diff --git a/deeppavlov/configs/classifiers/sentiment_sst_conv_bert.json b/deeppavlov/configs/classifiers/sentiment_sst_conv_bert.json
@@ -54,7 +54,6 @@
         "class_name": "torch_transformers_classifier",
         "n_classes": "#classes_vocab.len",
         "return_probas": true,
-        "one_hot_labels": true,
         "pretrained_bert": "{TRANSFORMER}",
         "save_path": "{MODEL_PATH}/model",
         "load_path": "{MODEL_PATH}/model",

diff --git a/deeppavlov/configs/classifiers/sentiment_twitter.json b/deeppavlov/configs/classifiers/sentiment_twitter.json
@@ -79,8 +79,7 @@
           "lr": 0.0001,
           "momentum": 0.9,
           "weight_decay": 0.0001
-        },
-        "model_name": "cnn_model"
+        }
       },
       {
         "in": "y_pred_probas",

diff --git a/deeppavlov/configs/classifiers/topics_distilbert_base_uncased.json b/deeppavlov/configs/classifiers/topics_distilbert_base_uncased.json
@@ -61,7 +61,6 @@
       },
       {
         "class_name": "torch_transformers_classifier",
-        "one_hot_labels": true,
         "n_classes": "#classes_vocab.len",
         "return_probas": true,
         "pretrained_bert": "{TRANSFORMER}",

diff --git a/deeppavlov/configs/entity_extraction/entity_linking_en.json b/deeppavlov/configs/entity_extraction/entity_linking_en.json
@@ -21,7 +21,6 @@
         "entity_ranker": "#entity_descr_ranking",
         "rank_in_runtime": true,
         "num_entities_for_bert_ranking": 20,
-        "use_gpu": false,
         "include_mention": false,
         "num_entities_to_return": 3,
         "lemmatize": true,

diff --git a/deeppavlov/configs/faq/fasttext_logreg.json b/deeppavlov/configs/faq/fasttext_logreg.json
@@ -104,7 +104,7 @@
         "subdir": "{DOWNLOADS_PATH}/massive/{LANGUAGE}"
       },
       {
-        "url": "https://files.deeppavlov.ai/faq/fasttext_logreg_{LANGUAGE}.tar.gz",
+        "url": "http://files.deeppavlov.ai/faq/fasttext_logreg_{LANGUAGE}.tar.gz",
         "subdir": "{MODEL_PATH}"
       }
     ]

diff --git a/deeppavlov/configs/ranking/path_ranking_nll_roberta_en.json b/deeppavlov/configs/ranking/path_ranking_nll_roberta_en.json
@@ -20,10 +20,10 @@
         "load_path": "{MODEL_PATH}/model",
         "encoder_save_path": "{MODEL_PATH}/encoder",
         "linear_save_path": "{MODEL_PATH}/linear",
-        "model_name": "in_batch_ranking_model",
         "pretrained_bert": "{TRANSFORMER}",
         "learning_rate_drop_patience": 5,
-        "learning_rate_drop_div": 1.5
+        "learning_rate_drop_div": 1.5,
+        "optimizer_parameters": {"lr": 1e-5, "weight_decay": 0.01, "eps": 1e-6}
       }
     ],
     "out": ["model_output"]

diff --git a/deeppavlov/configs/ranking/ranking_ubuntu_v2_torch_bert_uncased.json b/deeppavlov/configs/ranking/ranking_ubuntu_v2_torch_bert_uncased.json
@@ -29,7 +29,6 @@
       },
       {
         "class_name": "torch_bert_ranker",
-        "one_hot_labels": false,
         "pretrained_bert": "bert-base-uncased",
         "save_path": "{MODEL_PATH}/model",
         "load_path": "{MODEL_PATH}/model",
@@ -43,6 +42,7 @@
           ],
           "eps": 1e-6
         },
+        "clip_norm": 1.0,
         "in": [
           "bert_features"
         ],

diff --git a/deeppavlov/configs/ranking/rel_ranking_nll_bert_ru.json b/deeppavlov/configs/ranking/rel_ranking_nll_bert_ru.json
@@ -19,10 +19,10 @@
         "load_path": "{MODEL_PATH}/model",
         "encoder_save_path": "{MODEL_PATH}/encoder",
         "linear_save_path": "{MODEL_PATH}/linear",
-        "model_name": "in_batch_ranking_model",
         "pretrained_bert": "{TRANSFORMER}",
         "learning_rate_drop_patience": 4,
-        "learning_rate_drop_div": 1.5
+        "learning_rate_drop_div": 1.5,
+        "optimizer_parameters": {"lr": 1e-5, "weight_decay": 0.01, "eps": 1e-6}
       }
     ],
     "out": ["model_output"]

diff --git a/deeppavlov/configs/relation_extraction/re_docred.json b/deeppavlov/configs/relation_extraction/re_docred.json
@@ -27,7 +27,7 @@
         "out": ["model_output"],
         "save_path": "{MODEL_PATH}/model",
         "load_path": "{MODEL_PATH}/model",
-        "model_name": "re_model",
+        "optimizer_parameters": {"lr": 5e-5, "weight_decay": 0.01, "eps": 1e-6},
         "n_classes": 97,
         "num_ner_tags": 6,
         "pretrained_bert": "bert-base-cased",

diff --git a/deeppavlov/configs/relation_extraction/re_rured.json b/deeppavlov/configs/relation_extraction/re_rured.json
@@ -29,7 +29,7 @@
         "out": ["model_output"],
         "save_path": "{MODEL_PATH}/model",
         "load_path": "{MODEL_PATH}/model",
-        "model_name": "re_model",
+        "optimizer_parameters": {"lr": 5e-5, "weight_decay": 0.01, "eps": 1e-6},
         "n_classes": 30,
         "num_ner_tags": 29,
         "pretrained_bert": "{TRANSFORMER}"