Update README.md
Browse files
README.md
CHANGED
|
@@ -28,15 +28,15 @@ library_name: transformers
|
|
| 28 |
### 학습 방법
|
| 29 |
|
| 30 |
```python
|
|
|
|
|
|
|
| 31 |
|
| 32 |
from transformers import RobertaTokenizer, RobertaForMaskedLM
|
| 33 |
from transformers import AutoModel, AutoTokenizer
|
| 34 |
-
|
| 35 |
model = RobertaForMaskedLM.from_pretrained(base_model)
|
| 36 |
tokenizer = AutoTokenizer.from_pretrained(base_tokenizer)
|
| 37 |
|
| 38 |
from transformers import LineByLineTextDataset
|
| 39 |
-
|
| 40 |
dataset = LineByLineTextDataset(
|
| 41 |
tokenizer=tokenizer,
|
| 42 |
file_path=fpath_dataset,
|
|
@@ -44,13 +44,11 @@ dataset = LineByLineTextDataset(
|
|
| 44 |
)
|
| 45 |
|
| 46 |
from transformers import DataCollatorForLanguageModeling
|
| 47 |
-
|
| 48 |
data_collator = DataCollatorForLanguageModeling(
|
| 49 |
tokenizer=tokenizer, mlm=True, mlm_probability=0.15
|
| 50 |
)
|
| 51 |
|
| 52 |
from transformers import Trainer, TrainingArguments
|
| 53 |
-
|
| 54 |
training_args = TrainingArguments(
|
| 55 |
output_dir=output_dir,
|
| 56 |
overwrite_output_dir=True,
|
|
@@ -69,29 +67,6 @@ trainer = Trainer(
|
|
| 69 |
)
|
| 70 |
|
| 71 |
train_metrics = trainer.train()
|
| 72 |
-
|
| 73 |
-
|
| 74 |
trainer.save_model(output_dir)
|
| 75 |
trainer.push_to_hub()
|
| 76 |
```
|
| 77 |
-
|
| 78 |
-
### 학습용 configuration
|
| 79 |
-
|
| 80 |
-
- number of epochs
|
| 81 |
-
```bash
|
| 82 |
-
epochs = 50
|
| 83 |
-
```
|
| 84 |
-
|
| 85 |
-
- JSON file
|
| 86 |
-
```json
|
| 87 |
-
[
|
| 88 |
-
{'basemodel' : 'againeureka/klue_roberta_base_for_legal',
|
| 89 |
-
'basetokenizer' : 'klue/roberta-base',
|
| 90 |
-
'trainmodel' : 'againeureka/toulmin_classifier8_klue_roberta_base_retrained6',
|
| 91 |
-
'batchsize' : 92,
|
| 92 |
-
'epochs' : epochs,
|
| 93 |
-
'push_to_hub' : True,
|
| 94 |
-
'is_on' : True,
|
| 95 |
-
},
|
| 96 |
-
]
|
| 97 |
-
```
|
|
|
|
| 28 |
### 학습 방법
|
| 29 |
|
| 30 |
```python
|
| 31 |
+
base_model = 'klue/roberta-base'
|
| 32 |
+
base_tokenizer = 'klue/roberta-base'
|
| 33 |
|
| 34 |
from transformers import RobertaTokenizer, RobertaForMaskedLM
|
| 35 |
from transformers import AutoModel, AutoTokenizer
|
|
|
|
| 36 |
model = RobertaForMaskedLM.from_pretrained(base_model)
|
| 37 |
tokenizer = AutoTokenizer.from_pretrained(base_tokenizer)
|
| 38 |
|
| 39 |
from transformers import LineByLineTextDataset
|
|
|
|
| 40 |
dataset = LineByLineTextDataset(
|
| 41 |
tokenizer=tokenizer,
|
| 42 |
file_path=fpath_dataset,
|
|
|
|
| 44 |
)
|
| 45 |
|
| 46 |
from transformers import DataCollatorForLanguageModeling
|
|
|
|
| 47 |
data_collator = DataCollatorForLanguageModeling(
|
| 48 |
tokenizer=tokenizer, mlm=True, mlm_probability=0.15
|
| 49 |
)
|
| 50 |
|
| 51 |
from transformers import Trainer, TrainingArguments
|
|
|
|
| 52 |
training_args = TrainingArguments(
|
| 53 |
output_dir=output_dir,
|
| 54 |
overwrite_output_dir=True,
|
|
|
|
| 67 |
)
|
| 68 |
|
| 69 |
train_metrics = trainer.train()
|
|
|
|
|
|
|
| 70 |
trainer.save_model(output_dir)
|
| 71 |
trainer.push_to_hub()
|
| 72 |
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|