Replace tf squad with pytorch configs (#1539)

* replace tf squad with torch * fixes * infer in call method * update redundant configs * reformat, update config names in odqa * fix config names in tests * refactor * update configs parameters * update squad configs in tests * refactor * add score from 0 to 1 * fix docs * update tests * fix docs * Update squad_ru_bert.json * fix multi-gpu inference * fix multi-gpu inference * fix squad training * odqa fixes * refactor * refactor: import order * refactor: squad_preprocessor imports * refactor * update config names * remove redundant squad configs * update docs * fix: removed unexisting configs from test_quick_start * feat: squad aliases * loading in base class * refactor * fix torch model * remove lang * remove lang * refactor: removed redundand newlines in two configs * remove strict load flag * rename configs, add squad2 config * update docs, remove redundant config * feat: instafail * replace checkpoints with old transformers version with new checkpoints * rename configs * refactor: docs and redundant requirements * add sent_tokenize Co-authored-by: Fedor Ignatov <ignatov.fedor@gmail.com>
deeppavlov · May 5, 2022 · 602334d · 602334d
1 parent 163420b
commit 602334d
Show file tree

Hide file tree

Showing 63 changed files with 561 additions and 3,793 deletions.
diff --git a/deeppavlov/configs/classifiers/insults_kaggle_bert.json b/deeppavlov/configs/classifiers/insults_kaggle_bert.json
@@ -141,7 +141,7 @@
         "subdir": "{DOWNLOADS_PATH}"
       },
       {
-        "url": "http://files.deeppavlov.ai/deeppavlov_data/classifiers/insults_kaggle_torch_bert_v0.tar.gz",
+        "url": "http://files.deeppavlov.ai/deeppavlov_data/classifiers/insults_kaggle_torch_bert_v5.tar.gz",
         "subdir": "{MODELS_PATH}/classifiers"
       }
     ]

diff --git a/deeppavlov/configs/odqa/en_odqa_infer_wiki.json b/deeppavlov/configs/odqa/en_odqa_infer_wiki.json
@@ -31,7 +31,7 @@
       {
         "class_name": "logit_ranker",
         "batch_size": 64,
-        "squad_model": {"config_path": "{CONFIGS_PATH}/squad/multi_squad_noans_infer.json"},
+        "squad_model": {"config_path": "{CONFIGS_PATH}/squad/qa_squad2_bert.json"},
         "sort_noans": true,
         "in": ["chunks", "questions"],
         "out": ["answer", "answer_score", "answer_place"]

diff --git a/deeppavlov/configs/odqa/en_odqa_pop_infer_enwiki20180211.json b/deeppavlov/configs/odqa/en_odqa_pop_infer_enwiki20180211.json
@@ -49,7 +49,7 @@
         "class_name": "logit_ranker",
         "batch_size": 10,
         "squad_model": {
-          "config_path": "{CONFIGS_PATH}/squad/multi_squad_noans_infer.json"
+          "config_path": "{CONFIGS_PATH}/squad/qa_squad2_bert.json"
         },
         "sort_noans": true,
         "in": [
@@ -78,4 +78,4 @@
     "download": [
     ]
   }
-}
+}
diff --git a/deeppavlov/configs/odqa/ru_odqa_infer_wiki.json b/deeppavlov/configs/odqa/ru_odqa_infer_wiki.json
@@ -43,7 +43,7 @@
       {
         "class_name": "logit_ranker",
         "batch_size": 10,
-        "squad_model": {"config_path": "{CONFIGS_PATH}/squad/squad_ru.json"},
+        "squad_model": {"config_path": "{CONFIGS_PATH}/squad/squad_ru_bert.json"},
         "sort_noans": true,
         "in": [
           "chunks",

diff --git a/deeppavlov/configs/odqa/ru_odqa_infer_wiki_retr_noans.json b/deeppavlov/configs/odqa/ru_odqa_infer_wiki_retr_noans.json
@@ -43,7 +43,7 @@
       {
         "class_name": "logit_ranker",
         "batch_size": 10,
-        "squad_model": {"config_path": "{CONFIGS_PATH}/squad/multi_squad_ru_retr_noans.json"},
+        "squad_model": {"config_path": "{CONFIGS_PATH}/squad/qa_multisberquad_bert.json"},
         "sort_noans": true,
         "in": [
           "chunks",

diff --git a/deeppavlov/configs/odqa/ru_odqa_infer_wiki_rubert.json b/deeppavlov/configs/odqa/ru_odqa_infer_wiki_rubert.json
diff --git a/deeppavlov/configs/odqa/ru_odqa_infer_wiki_rubert_noans.json b/deeppavlov/configs/odqa/ru_odqa_infer_wiki_rubert_noans.json
diff --git a/deeppavlov/configs/ranking/ranking_ubuntu_v2_torch_bert_uncased.json b/deeppavlov/configs/ranking/ranking_ubuntu_v2_torch_bert_uncased.json
@@ -89,7 +89,7 @@
         "subdir": "{DOWNLOADS_PATH}/ubuntu_v2_data"
       },
       {
-        "url": "http://files.deeppavlov.ai/deeppavlov_data/ubuntu_v2_uncased_torch_bert_model_v0.tar.gz",
+        "url": "http://files.deeppavlov.ai/deeppavlov_data/ubuntu_v2_uncased_torch_bert_model_v2.tar.gz",
         "subdir": "{MODELS_PATH}"
       }
     ]

diff --git a/deeppavlov/configs/relation_extraction/re_docred.json b/deeppavlov/configs/relation_extraction/re_docred.json
@@ -17,7 +17,7 @@
         "in": ["tokens", "entity_pos", "entity_tags"],
         "out": ["input_ids", "attention_mask", "upd_entity_pos", "upd_entity_tags", "nf_samples"],
         "class_name": "re_preprocessor",
-        "vocab_file": "bert-base-uncased",
+        "vocab_file": "bert-base-cased",
         "default_tag": "PER"
       },
       {
@@ -30,7 +30,7 @@
         "model_name": "re_model",
         "n_classes": 97,
         "num_ner_tags": 6,
-        "pretrained_bert": "bert-base-uncased",
+        "pretrained_bert": "bert-base-cased",
         "return_probas": true
       },
       {
@@ -45,13 +45,13 @@
   },
   "train": {
     "epochs": 50,
-    "batch_size": 16,
+    "batch_size": 30,
     "log_every_n_batches": 100,
     "train_metrics": ["f1_weighted", "acc"],
     "evaluation_targets": ["valid", "train"],
     "metrics": ["f1_weighted", "acc"],
     "validation_patience": 50,
-    "val_every_n_batches": 100,
+    "val_every_n_batches": 200,
     "show_examples": false,
     "class_name": "torch_trainer"
   },
@@ -68,7 +68,7 @@
         "subdir": "{DOWNLOADS_PATH}/docred"
       },
       {
-        "url": "http://files.deeppavlov.ai/deeppavlov_data/relation_extraction/re_docred_model.tar.gz",
+        "url": "http://files.deeppavlov.ai/deeppavlov_data/relation_extraction/re_docred_model_v1.tar.gz",
         "subdir": "{MODELS_PATH}/re_docred"
       },
       {

diff --git a/deeppavlov/configs/relation_extraction/re_rured.json b/deeppavlov/configs/relation_extraction/re_rured.json
@@ -70,7 +70,7 @@
         "subdir": "{DOWNLOADS_PATH}/rured"
       },
       {
-        "url": "http://files.deeppavlov.ai/deeppavlov_data/relation_extraction/re_rured_model.tar.gz",
+        "url": "http://files.deeppavlov.ai/deeppavlov_data/relation_extraction/re_rured_model_v1.tar.gz",
         "subdir": "{MODELS_PATH}/re_rured"
       }
     ]