kaiser/withProto0110.out at master · machinereading/kaiser · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
Using TensorFlow backend.
### Korean FrameNet ###
	# contact: hahmyg@kaist, hahmyg@gmail.com #


withProto
### TRAINING
MODEL: framenet
LANGUAGE: multi
PRETRAINED BERT: bert-base-multilingual-cased
training data:
	(ko): 1783
BATCH_SIZE: 6
MAX_LEN: 256

used dictionary:
	 /disk/kaiser/kaiser/src/../koreanframenet/resource/info/mul_lu2idx.json
	 /disk/kaiser/kaiser/src/../koreanframenet/resource/info/mul_lufrmap.json
	 /disk/kaiser/kaiser/src/../koreanframenet/resource/info/mul_bio_frargmap.json
original model: /disk/data/models/frameBERT/frameBERT_en/

	your model would be saved at /disk/data/models/framenet/withProto-100/
retrain: True

### converting data to BERT input...
	 ...is done: 0hour:0min:2sec
	#of instance: 1783 1783
Epoch:   0%|                                                | 0/50 [00:00<?, ?it/s]../kaiser/src/utils.py:309: UserWarning: Implicit dimension choice for softmax has been deprecated. Change the call to include dim=X as an argument.
  pred_logits = sm(masked_logit).view(1,-1)
Train loss: 3.634308519779436

	your model is saved: /disk/data/models/framenet/withProto-100/0/
Epoch:   2%|▋                                    | 1/50 [01:52<1:32:03, 112.73s/it]Train loss: 3.409669096837908

	your model is saved: /disk/data/models/framenet/withProto-100/1/
Epoch:   4%|█▍                                   | 2/50 [04:00<1:33:50, 117.31s/it]Train loss: 3.278326953817534

	your model is saved: /disk/data/models/framenet/withProto-100/2/
Epoch:   6%|██▏                                  | 3/50 [06:09<1:34:38, 120.82s/it]