entity linking to pytorch and reduce ram and gpu (#1516)

* lite entity linking * fixes * kbqa to pytorch and lite * el and kbqa updates * remove redundant components * update outputs of ner configs * update components registry * reformat * docstrings * rename configs * update tests * update docs * update docs * add init.py * refactor * fix tests * fix tests * fix requirements * fixes * fix kbqa requirements and downloads * fix downloads * update requirements * fix tests * fix downloads * feat: kbqa aliases * remove: rel_ranking_infer requirement_registry duplicate * refactor * refactor * update typing * refactor * refactor * fixes * refactor: license, import order, etc. * refactor: removed ranking training configs, devices type changed Co-authored-by: Fedor Ignatov <ignatov.fedor@gmail.com>
deeppavlov · May 4, 2022 · 7f1c926 · 7f1c926
1 parent abbd090
commit 7f1c926
Show file tree

Hide file tree

Showing 63 changed files with 3,418 additions and 3,854 deletions.
diff --git a/deeppavlov/configs/classifiers/entity_ranking_bert_eng_no_mention.json b/deeppavlov/configs/classifiers/entity_ranking_bert_eng_no_mention.json
diff --git a/deeppavlov/configs/classifiers/entity_ranking_bert_rus_no_mention.json b/deeppavlov/configs/classifiers/entity_ranking_bert_rus_no_mention.json
diff --git a/deeppavlov/configs/classifiers/query_pr.json b/deeppavlov/configs/classifiers/query_pr.json
@@ -1,9 +1,7 @@
 {
   "dataset_reader": {
-    "class_name": "basic_classification_reader",
-    "x": "Question",
-    "y": "Class",
-    "data_path": "{DOWNLOADS_PATH}/query_prediction"
+    "class_name": "sq_reader",
+    "data_path": "{DOWNLOADS_PATH}/query_prediction/query_prediction_eng.pickle"
   },
   "dataset_iterator": {
     "class_name": "basic_classification_iterator",
@@ -14,8 +12,8 @@
     "in_y": ["y"],
     "pipe": [
       {
-        "class_name": "bert_preprocessor",
-        "vocab_file": "{BERT_PATH}/vocab.txt",
+        "class_name": "torch_transformers_preprocessor",
+        "vocab_file": "{TRANSFORMER}",
         "do_lower_case": false,
         "max_seq_length": 64,
         "in": ["x"],
@@ -27,42 +25,40 @@
         "fit_on": ["y"],
         "save_path": "{MODEL_PATH}/classes.dict",
         "load_path": "{MODEL_PATH}/classes.dict",
-        "in": "y",
-        "out": "y_ids"
+        "in": ["y"],
+        "out": ["y_ids"]
       },
       {
-        "in": "y_ids",
-        "out": "y_onehot",
+        "in": ["y_ids"],
+        "out": ["y_onehot"],
         "class_name": "one_hotter",
         "depth": "#classes_vocab.len",
         "single_vector": true
       },
       {
-        "class_name": "bert_classifier",
+        "class_name": "torch_transformers_classifier",
         "n_classes": "#classes_vocab.len",
         "return_probas": true,
-        "one_hot_labels": true,
-        "bert_config_file": "{BERT_PATH}/bert_config.json",
-        "pretrained_bert": "{BERT_PATH}/bert_model.ckpt",
+        "pretrained_bert": "{TRANSFORMER}",
         "save_path": "{MODEL_PATH}/model",
         "load_path": "{MODEL_PATH}/model",
-        "keep_prob": 0.5,
-        "learning_rate": 1e-05,
+        "optimizer": "AdamW",
+        "optimizer_parameters": {"lr": 1e-05},
         "learning_rate_drop_patience": 5,
         "learning_rate_drop_div": 2.0,
         "in": ["bert_features"],
-        "in_y": ["y_onehot"],
+        "in_y": ["y_ids"],
         "out": ["y_pred_probas"]
       },
       {
-        "in": "y_pred_probas",
-        "out": "y_pred_ids",
+        "in": ["y_pred_probas"],
+        "out": ["y_pred_ids"],
         "class_name": "proba2labels",
         "max_proba": true
       },
       {
-        "in": "y_pred_ids",
-        "out": "y_pred_labels",
+        "in": ["y_pred_ids"],
+        "out": ["y_pred_labels"],
         "ref": "classes_vocab"
       }
     ],
@@ -72,44 +68,36 @@
     "epochs": 100,
     "batch_size": 64,
     "metrics": [
-      "sets_accuracy",
       "f1_macro",
+      "accuracy",
       {
         "name": "roc_auc",
         "inputs": ["y_onehot", "y_pred_probas"]
       }
     ],
-    "validation_patience": 5,
-    "val_every_n_epochs": 1,
-    "log_every_n_epochs": 1,
+    "validation_patience": 10,
+    "val_every_n_batches": 100,
+    "log_every_n_batches": 100,
     "show_examples": false,
     "evaluation_targets": ["train", "valid", "test"],
-    "class_name": "nn_trainer",
-    "tensorboard_log_dir": "{MODEL_PATH}/"
+    "class_name": "torch_trainer"
   },
   "metadata": {
     "variables": {
+      "TRANSFORMER": "haisongzhang/roberta-tiny-cased",
       "ROOT_PATH": "~/.deeppavlov",
       "DOWNLOADS_PATH": "{ROOT_PATH}/downloads",
       "MODELS_PATH": "{ROOT_PATH}/models",
-      "BERT_PATH": "{DOWNLOADS_PATH}/bert_models_kbqa/cased_L-12_H-768_A-12",
-      "MODEL_PATH": "{MODELS_PATH}/classifiers/query_prediction"
-    },
-    "labels": {
-      "server_utils": "KerasIntentModel"
+      "MODEL_PATH": "{MODELS_PATH}/classifiers/query_prediction_eng"
     },
     "download": [
       {
-        "url": "http://files.deeppavlov.ai/kbqa/datasets/query_prediction.tar.gz",
-        "subdir": "{DOWNLOADS_PATH}/query_prediction"
+        "url": "http://files.deeppavlov.ai/kbqa/wikidata/query_prediction_eng.tar.gz",
+        "subdir": "{MODELS_PATH}/classifiers/query_prediction_eng"
       },
       {
-        "url": "http://files.deeppavlov.ai/deeppavlov_data/bert/cased_L-12_H-768_A-12.zip",
-        "subdir": "{DOWNLOADS_PATH}/bert_models_kbqa"
-      },
-      {
-        "url": "http://files.deeppavlov.ai/kbqa/models/query_prediction.tar.gz",
-        "subdir": "{MODELS_PATH}/classifiers/query_prediction"
+        "url": "http://files.deeppavlov.ai/kbqa/wikidata/query_prediction_eng.pickle",
+        "subdir": "{DOWNLOADS_PATH}/query_prediction"
       }
     ]
   }

diff --git a/deeppavlov/configs/classifiers/rel_ranking_bert.json b/deeppavlov/configs/classifiers/rel_ranking_bert.json