Skip to content

Commit 0448c73

Browse files
IgnatovFedorIgnatovFedor
authored
Release 1.1.1
2 parents 8d8bdaf + 3206205 commit 0448c73

28 files changed

+2023
-108
lines changed

‎deeppavlov/_meta.py

+1-1
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,4 @@
1-
__version__ = '1.1.0'
1+
__version__ = '1.1.1'
22
__author__ = 'Neural Networks and Deep Learning lab, MIPT'
33
__description__ = 'An open source library for building end-to-end dialog systems and training chatbots.'
44
__keywords__ = ['NLP', 'NER', 'SQUAD', 'Intents', 'Chatbot']

‎deeppavlov/configs/multitask/mt_glue.json

+287
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,287 @@
1+
{
2+
"dataset_reader": {
3+
"class_name": "multitask_reader",
4+
"task_defaults": {
5+
"class_name": "huggingface_dataset_reader",
6+
"path": "glue",
7+
"train": "train",
8+
"valid": "validation"
9+
},
10+
"tasks": {
11+
"cola": {"name": "cola"},
12+
"sst2": {"name": "sst2"},
13+
"qqp": {"name": "qqp"},
14+
"mrpc": {"name": "mrpc"},
15+
"rte": {"name": "rte"},
16+
"mnli": {
17+
"name": "mnli",
18+
"valid": "validation_matched"
19+
},
20+
"qnli": {"name": "qnli"},
21+
"stsb": {"name": "stsb"}
22+
}
23+
},
24+
"dataset_iterator": {
25+
"class_name": "multitask_iterator",
26+
"num_train_epochs": "{NUM_TRAIN_EPOCHS}",
27+
"gradient_accumulation_steps": "{GRADIENT_ACC_STEPS}",
28+
"seed": 42,
29+
"task_defaults": {
30+
"class_name": "huggingface_dataset_iterator",
31+
"label": "label",
32+
"use_label_name": false,
33+
"seed": 42
34+
},
35+
"tasks": {
36+
"cola": {
37+
"features": ["sentence"]
38+
},
39+
"sst2": {
40+
"features": ["sentence"]
41+
},
42+
"qqp": {
43+
"features": ["question1", "question2"]
44+
},
45+
"mrpc": {
46+
"features": ["sentence1", "sentence2"]
47+
},
48+
"rte": {
49+
"features": ["sentence1", "sentence2"]
50+
},
51+
"mnli": {
52+
"features": ["premise", "hypothesis"]
53+
},
54+
"qnli": {
55+
"features": ["question", "sentence"]
56+
},
57+
"stsb": {
58+
"features": ["sentence1", "sentence2"]
59+
}
60+
}
61+
},
62+
"chainer": {
63+
"in": ["x_cola", "x_sst2", "x_qqp", "x_mrpc", "x_rte", "x_mnli", "x_qnli", "x_stsb"],
64+
"in_y": ["y_cola", "y_sst2", "y_qqp", "y_mrpc", "y_rte", "y_mnli", "y_qnli", "y_stsb"
65+
],
66+
"pipe": [
67+
{
68+
"class_name": "multitask_pipeline_preprocessor",
69+
"possible_keys_to_extract": [0, 1],
70+
"preprocessor": "TorchTransformersPreprocessor",
71+
"vocab_file": "{BACKBONE}",
72+
"max_seq_length": 128,
73+
"do_lower_case": true,
74+
"n_task": 8,
75+
"in": ["x_cola", "x_sst2", "x_qqp", "x_mrpc", "x_rte", "x_mnli", "x_qnli", "x_stsb"],
76+
"out": [
77+
"bert_features_cola",
78+
"bert_features_sst2",
79+
"bert_features_qqp",
80+
"bert_features_mrpc",
81+
"bert_features_rte",
82+
"bert_features_mnli",
83+
"bert_features_qnli",
84+
"bert_features_stsb"
85+
]
86+
},
87+
{
88+
"id": "multitask_transformer",
89+
"class_name": "multitask_transformer",
90+
"optimizer_parameters": {"lr": 2e-5},
91+
"gradient_accumulation_steps": "{GRADIENT_ACC_STEPS}",
92+
"learning_rate_drop_patience": 2,
93+
"learning_rate_drop_div": 2.0,
94+
"return_probas": true,
95+
"backbone_model": "{BACKBONE}",
96+
"save_path": "{MODEL_PATH}",
97+
"load_path": "{MODEL_PATH}",
98+
"tasks": {
99+
"cola": {
100+
"type": "classification",
101+
"options": 2
102+
},
103+
"sst2": {
104+
"type": "classification",
105+
"options": 2
106+
},
107+
"qqp": {
108+
"type": "classification",
109+
"options": 2
110+
},
111+
"mrpc": {
112+
"type": "classification",
113+
"options": 2
114+
},
115+
"rte": {
116+
"type": "classification",
117+
"options": 2
118+
},
119+
"mnli": {
120+
"type": "classification",
121+
"options": 3
122+
},
123+
"qnli": {
124+
"type": "classification",
125+
"options": 2
126+
},
127+
"stsb": {
128+
"type": "regression",
129+
"options": 1
130+
}
131+
},
132+
"in": [
133+
"bert_features_cola",
134+
"bert_features_sst2",
135+
"bert_features_qqp",
136+
"bert_features_mrpc",
137+
"bert_features_rte",
138+
"bert_features_mnli",
139+
"bert_features_qnli",
140+
"bert_features_stsb"
141+
],
142+
"in_y": ["y_cola", "y_sst2", "y_qqp", "y_mrpc", "y_rte", "y_mnli", "y_qnli", "y_stsb"],
143+
"out": [
144+
"y_cola_pred_probas",
145+
"y_sst2_pred_probas",
146+
"y_qqp_pred_probas",
147+
"y_mrpc_pred_probas",
148+
"y_rte_pred_probas",
149+
"y_mnli_pred_probas",
150+
"y_qnli_pred_probas",
151+
"y_stsb_pred"
152+
]
153+
},
154+
{
155+
"in": [
156+
"y_cola_pred_probas",
157+
"y_sst2_pred_probas",
158+
"y_qqp_pred_probas",
159+
"y_mrpc_pred_probas",
160+
"y_rte_pred_probas",
161+
"y_mnli_pred_probas",
162+
"y_qnli_pred_probas"
163+
],
164+
"out": [
165+
"y_cola_pred_ids",
166+
"y_sst2_pred_ids",
167+
"y_qqp_pred_ids",
168+
"y_mrpc_pred_ids",
169+
"y_rte_pred_ids",
170+
"y_mnli_pred_ids",
171+
"y_qnli_pred_ids"
172+
],
173+
"class_name": "proba2labels",
174+
"max_proba": true
175+
}
176+
],
177+
"out": [
178+
"y_cola_pred_probas",
179+
"y_sst2_pred_probas",
180+
"y_qqp_pred_probas",
181+
"y_mrpc_pred_probas",
182+
"y_rte_pred_probas",
183+
"y_mnli_pred_probas",
184+
"y_qnli_pred_probas",
185+
"y_stsb_pred",
186+
"y_cola_pred_ids",
187+
"y_sst2_pred_ids",
188+
"y_qqp_pred_ids",
189+
"y_mrpc_pred_ids",
190+
"y_rte_pred_ids",
191+
"y_mnli_pred_ids",
192+
"y_qnli_pred_ids",
193+
"y_stsb_pred"
194+
]
195+
},
196+
"train": {
197+
"epochs": "{NUM_TRAIN_EPOCHS}",
198+
"batch_size": 32,
199+
"metrics": [
200+
{
201+
"name": "multitask_accuracy",
202+
"inputs": [
203+
"y_rte",
204+
"y_mnli",
205+
"y_qnli",
206+
"y_mrpc",
207+
"y_cola",
208+
"y_sst2",
209+
"y_qqp",
210+
"y_rte_pred_ids",
211+
"y_mnli_pred_ids",
212+
"y_qnli_pred_ids",
213+
"y_mrpc_pred_ids",
214+
"y_cola_pred_ids",
215+
"y_sst2_pred_ids",
216+
"y_qqp_pred_ids"
217+
]
218+
},
219+
{
220+
"name": "accuracy",
221+
"alias": "accuracy_mrpc",
222+
"inputs": ["y_mrpc", "y_mrpc_pred_ids"]
223+
},
224+
{
225+
"name": "accuracy",
226+
"alias": "accuracy_rte",
227+
"inputs": ["y_rte", "y_rte_pred_ids"]
228+
},
229+
{
230+
"name": "accuracy",
231+
"alias": "accuracy_mnli",
232+
"inputs": ["y_mnli", "y_mnli_pred_ids"]
233+
},
234+
{
235+
"name": "accuracy",
236+
"alias": "accuracy_qnli",
237+
"inputs": ["y_qnli", "y_qnli_pred_ids"]
238+
},
239+
{
240+
"name": "accuracy",
241+
"alias": "accuracy_sst",
242+
"inputs": ["y_sst2", "y_sst2_pred_ids"]
243+
},
244+
{
245+
"name": "accuracy",
246+
"alias": "accuracy_cola",
247+
"inputs": ["y_cola", "y_cola_pred_ids"]
248+
},
249+
{
250+
"name": "accuracy",
251+
"alias": "accuracy_qqp",
252+
"inputs": ["y_qqp", "y_qqp_pred_ids"]
253+
},
254+
{
255+
"name": "pearson_correlation",
256+
"alias": "pearson_correlation_stsb",
257+
"inputs": ["y_stsb", "y_stsb_pred"]
258+
},
259+
{
260+
"name": "spearman_correlation",
261+
"alias": "spearman_correlation_stsb",
262+
"inputs": ["y_stsb", "y_stsb_pred"]
263+
}
264+
],
265+
"validation_patience": 3,
266+
"val_every_n_epochs": 1,
267+
"log_every_n_epochs": 1,
268+
"show_examples": false,
269+
"evaluation_targets": ["valid"],
270+
"class_name": "torch_trainer"
271+
},
272+
"metadata": {
273+
"variables": {
274+
"BACKBONE": "bert-base-uncased",
275+
"MODELS_PATH": "~/.deeppavlov/models/glue",
276+
"MODEL_PATH": "{MODELS_PATH}/8task",
277+
"NUM_TRAIN_EPOCHS": 5,
278+
"GRADIENT_ACC_STEPS": 1
279+
},
280+
"download": [
281+
{
282+
"url": "http://files.deeppavlov.ai/deeppavlov_data/multitask/glue.tar.gz",
283+
"subdir": "{MODELS_PATH}"
284+
}
285+
]
286+
}
287+
}

0 commit comments

Comments
 (0)