refactor: squad_torch_bert[_infer] -> squad_bert[_infer]

deeppavlov · Feb 1, 2022 · fb5bfed · fb5bfed
1 parent 40d6e80
commit fb5bfed
Show file tree

Hide file tree

Showing 7 changed files with 128 additions and 322 deletions.
diff --git a/deeppavlov/configs/squad/squad_bert.json b/deeppavlov/configs/squad/squad_bert.json
@@ -9,55 +9,115 @@
     "shuffle": true
   },
   "chainer": {
-    "in": ["context_raw", "question_raw"],
-    "in_y": ["ans_raw", "ans_raw_start"],
+    "in": [
+      "context_raw",
+      "question_raw"
+    ],
+    "in_y": [
+      "ans_raw",
+      "ans_raw_start"
+    ],
     "pipe": [
       {
-        "class_name": "bert_preprocessor",
-        "vocab_file": "{DOWNLOADS_PATH}/bert_models/cased_L-12_H-768_A-12/vocab.txt",
-        "do_lower_case": false,
+        "class_name": "torch_squad_transformers_preprocessor",
+        "vocab_file": "{TRANSFORMER}",
+        "do_lower_case": "{LOWERCASE}",
         "max_seq_length": 384,
-        "in": ["question_raw", "context_raw"],
-        "out": ["bert_features"]
+        "return_tokens": true,
+        "in": [
+          "question_raw",
+          "context_raw"
+        ],
+        "out": [
+          "bert_features",
+          "subtokens"
+        ]
       },
       {
         "class_name": "squad_bert_mapping",
-        "do_lower_case": false,
-        "in": ["context_raw", "bert_features"],
-        "out": ["subtok2chars", "char2subtoks"]
+        "do_lower_case": "{LOWERCASE}",
+        "in": [
+          "context_raw",
+          "bert_features",
+          "subtokens"
+        ],
+        "out": [
+          "subtok2chars",
+          "char2subtoks"
+        ]
       },
       {
         "class_name": "squad_bert_ans_preprocessor",
-        "do_lower_case": false,
-        "in": ["ans_raw", "ans_raw_start","char2subtoks"],
-        "out": ["ans", "ans_start", "ans_end"]
+        "do_lower_case": "{LOWERCASE}",
+        "in": [
+          "ans_raw",
+          "ans_raw_start",
+          "char2subtoks"
+        ],
+        "out": [
+          "ans",
+          "ans_start",
+          "ans_end"
+        ]
       },
       {
-        "class_name": "squad_bert_model",
-        "bert_config_file": "{DOWNLOADS_PATH}/bert_models/cased_L-12_H-768_A-12/bert_config.json",
-        "pretrained_bert": "{DOWNLOADS_PATH}/bert_models/cased_L-12_H-768_A-12/bert_model.ckpt",
-        "save_path": "{MODELS_PATH}/squad_bert/model",
-        "load_path": "{MODELS_PATH}/squad_bert/model",
-        "keep_prob": 0.5,
-        "learning_rate": 2e-05,
+        "class_name": "torch_transformers_squad",
+        "pretrained_bert": "{TRANSFORMER}",
+        "save_path": "{MODEL_PATH}/model",
+        "load_path": "{MODEL_PATH}/model",
+        "optimizer": "AdamW",
+        "optimizer_parameters": {
+          "lr": 2e-05,
+          "weight_decay": 0.01,
+          "betas": [
+            0.9,
+            0.999
+          ],
+          "eps": 1e-06
+        },
         "learning_rate_drop_patience": 2,
         "learning_rate_drop_div": 2.0,
-        "in": ["bert_features"],
-        "in_y": ["ans_start", "ans_end"],
-        "out": ["ans_start_predicted", "ans_end_predicted", "logits"]
+        "in": [
+          "bert_features"
+        ],
+        "in_y": [
+          "ans_start",
+          "ans_end"
+        ],
+        "out": [
+          "ans_start_predicted",
+          "ans_end_predicted",
+          "logits"
+        ]
       },
       {
         "class_name": "squad_bert_ans_postprocessor",
-        "in": ["ans_start_predicted", "ans_end_predicted", "context_raw", "bert_features", "subtok2chars"],
-        "out": ["ans_predicted", "ans_start_predicted", "ans_end_predicted"]
+        "in": [
+          "ans_start_predicted",
+          "ans_end_predicted",
+          "context_raw",
+          "bert_features",
+          "subtok2chars",
+          "subtokens"
+        ],
+        "out": [
+          "ans_predicted",
+          "ans_start_predicted",
+          "ans_end_predicted"
+        ]
       }
     ],
-    "out": ["ans_predicted", "ans_start_predicted", "logits"]
+    "out": [
+      "ans_predicted",
+      "ans_start_predicted",
+      "logits"
+    ]
   },
   "train": {
     "show_examples": false,
-    "test_best": false,
-    "validate_best": true,
+    "evaluation_targets": [
+      "valid"
+    ],
     "log_every_n_batches": 250,
     "val_every_n_batches": 500,
     "batch_size": 10,
@@ -67,39 +127,49 @@
     "metrics": [
       {
         "name": "squad_v1_f1",
-        "inputs": ["ans", "ans_predicted"]
+        "inputs": [
+          "ans",
+          "ans_predicted"
+        ]
       },
       {
         "name": "squad_v1_em",
-        "inputs": ["ans", "ans_predicted"]
+        "inputs": [
+          "ans",
+          "ans_predicted"
+        ]
       },
       {
         "name": "squad_v2_f1",
-        "inputs": ["ans", "ans_predicted"]
+        "inputs": [
+          "ans",
+          "ans_predicted"
+        ]
       },
       {
         "name": "squad_v2_em",
-        "inputs": ["ans", "ans_predicted"]
+        "inputs": [
+          "ans",
+          "ans_predicted"
+        ]
       }
     ],
-    "tensorboard_log_dir": "{MODELS_PATH}/squad_bert/logs"
+    "class_name": "torch_trainer"
   },
   "metadata": {
     "variables": {
+      "LOWERCASE": true,
+      "TRANSFORMER": "bert-base-uncased",
       "ROOT_PATH": "~/.deeppavlov",
       "DOWNLOADS_PATH": "{ROOT_PATH}/downloads",
-      "MODELS_PATH": "{ROOT_PATH}/models"
+      "MODELS_PATH": "{ROOT_PATH}/models",
+      "MODEL_PATH": "{MODELS_PATH}/squad_torch_bert/{TRANSFORMER}"
     },
     "download": [
       {
-        "url": "http://files.deeppavlov.ai/deeppavlov_data/bert/cased_L-12_H-768_A-12.zip",
-        "subdir": "{DOWNLOADS_PATH}/bert_models"
-      },
-      {
-        "url": "http://files.deeppavlov.ai/deeppavlov_data/squad_bert.tar.gz",
-        "subdir": "{MODELS_PATH}"
+        "url": "http://files.deeppavlov.ai/v1/squad/squad_torch_bert.tar.gz",
+        "subdir": "{ROOT_PATH}/models"
       }
-      ]
+    ]
   }
 }
-
diff --git a/deeppavlov/configs/squad/squad_bert_infer.json b/deeppavlov/configs/squad/squad_bert_infer.json
@@ -13,12 +13,12 @@
     "in_y": ["ans_raw", "ans_raw_start"],
     "pipe": [
       {
-        "class_name": "squad_bert_infer",
+        "class_name": "torch_transformers_squad_infer",
         "batch_size": 10,
         "squad_model_config": "{CONFIGS_PATH}/squad/squad_bert.json",
-        "vocab_file": "{DOWNLOADS_PATH}/bert_models/cased_L-12_H-768_A-12/vocab.txt",
+        "vocab_file": "bert-base-cased",
         "do_lower_case": false,
-        "max_seq_length": 512,
+        "max_seq_length": 384,
         "in": ["context_raw", "question_raw"],
         "out": ["ans_predicted", "ans_start_predicted", "logits"]
       }
@@ -27,8 +27,9 @@
   },
   "train": {
     "show_examples": false,
-    "test_best": false,
-    "validate_best": true,
+    "evaluation_targets": [
+      "valid"
+    ],
     "log_every_n_batches": 250,
     "val_every_n_batches": 500,
     "batch_size": 10,
@@ -51,24 +52,17 @@
         "name": "squad_v2_em",
         "inputs": ["ans_raw", "ans_predicted"]
       }
-    ],
-    "tensorboard_log_dir": "{MODELS_PATH}/squad_bert/logs"
+    ]
   },
   "metadata": {
     "variables": {
       "ROOT_PATH": "~/.deeppavlov",
+      "TRANSFORMER": "bert-base-cased",
       "DOWNLOADS_PATH": "{ROOT_PATH}/downloads",
-      "MODELS_PATH": "{ROOT_PATH}/models",
+      "MODELS_PATH": "{ROOT_PATH}/models/squad_torch_bert/{TRANSFORMER}",
       "CONFIGS_PATH": "{DEEPPAVLOV_PATH}/configs"
     },
-    "download": [{
-	    "url": "http://files.deeppavlov.ai/deeppavlov_data/bert/cased_L-12_H-768_A-12.zip",
-	    "subdir": "{DOWNLOADS_PATH}/bert_models"
-    },
-    {
-	    "url": "http://files.deeppavlov.ai/deeppavlov_data/squad_bert.tar.gz",
-	    "subdir": "{MODELS_PATH}"
-    }
+    "download": [
     ]
   }
 }