Commit
•
e2a78ee
1
Parent(s):
7d05d99
Update: 리드미 완전 추가
Browse files
README.md
CHANGED
@@ -16,8 +16,8 @@ model-index:
|
|
16 |
- name: barTNumText
|
17 |
results:
|
18 |
- task:
|
19 |
-
type:
|
20 |
-
name:
|
21 |
metrics:
|
22 |
- type: bleu # Required. Example: wer. Use metric id from https://hf.co/metrics
|
23 |
value: 0.9161441917016176 # Required. Example: 20.90
|
@@ -39,4 +39,88 @@ model-index:
|
|
39 |
value: 0.9500390902948073 # Required. Example: 20.90
|
40 |
name: eval_rougeLsum # Optional. Example: Test WER
|
41 |
verified: true # Optional. If true, indicates that evaluation was generated by Hugging Face (vs. self-reported).
|
42 |
-
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
16 |
- name: barTNumText
|
17 |
results:
|
18 |
- task:
|
19 |
+
type: text2text-generation # Required. Example: automatic-speech-recognition
|
20 |
+
name: text2text-generation # Optional. Example: Speech Recognition
|
21 |
metrics:
|
22 |
- type: bleu # Required. Example: wer. Use metric id from https://hf.co/metrics
|
23 |
value: 0.9161441917016176 # Required. Example: 20.90
|
|
|
39 |
value: 0.9500390902948073 # Required. Example: 20.90
|
40 |
name: eval_rougeLsum # Optional. Example: Test WER
|
41 |
verified: true # Optional. If true, indicates that evaluation was generated by Hugging Face (vs. self-reported).
|
42 |
+
---
|
43 |
+
|
44 |
+
# CamemBERT: a Tasty French Language Model
|
45 |
+
|
46 |
+
## Table of Contents
|
47 |
+
- [CamemBERT: a Tasty French Language Model](#camembert-a-tasty-french-language-model)
|
48 |
+
- [Table of Contents](#table-of-contents)
|
49 |
+
- [Model Details](#model-details)
|
50 |
+
- [Uses](#uses)
|
51 |
+
- [Evaluation](#evaluation)
|
52 |
+
- [How to Get Started With the Model](#how-to-get-started-with-the-model)
|
53 |
+
|
54 |
+
|
55 |
+
## Model Details
|
56 |
+
- **Model Description:**
|
57 |
+
뭔가 찾아봐도 모델이나 알고리즘이 딱히 없어서 만들어본 모델입니다. <br />
|
58 |
+
BartForConditionalGeneration Fine-Tuning Model For Number To Korean <br />
|
59 |
+
BartForConditionalGeneration으로 파인튜닝한, 숫자를 한글로 변환하는 Task 입니다. <br />
|
60 |
+
|
61 |
+
Dataset use [Korea aihub](https://aihub.or.kr/aihubdata/data/list.do?currMenu=115&topMenu=100&srchDataRealmCode=REALM002&srchDataTy=DATA004) <br />
|
62 |
+
I can't open my fine-tuning datasets for my private issue <br />
|
63 |
+
데이터셋은 Korea aihub에서 받아서 사용하였으며, 파인튜닝에 사용된 모든 데이터를 사정상 공개해드릴 수는 없습니다. <br />
|
64 |
+
|
65 |
+
Korea aihub data is ONLY permit to Korean!!!!!!! <br />
|
66 |
+
aihub에서 데이터를 받으실 분은 한국인일 것이므로, 한글로만 작성합니다. <br />
|
67 |
+
정확히는 음성전사를 철자전사로 번역하는 형태로 학습된 모델입니다. (ETRI 전사기준) <br />
|
68 |
+
|
69 |
+
In case, ten million, some people use 10 million or some people use 10000000, so this model is crucial for training datasets
|
70 |
+
천만을 1000만 혹은 10000000으로 쓸 수도 있기에, Training Datasets에 따라 결과는 상이할 수 있습니다. <br />
|
71 |
+
- **Developed by:** Yoo SungHyun(https://github.com/YooSungHyun)
|
72 |
+
- **Language(s):** Korean
|
73 |
+
- **License:** apache-2.0
|
74 |
+
- **Parent Model:** See the [kobart-base-v2](https://huggingface.co/gogamza/kobart-base-v2) for more information about the pre-trained base model.
|
75 |
+
|
76 |
+
|
77 |
+
## Uses
|
78 |
+
This Model is inferenced token BACKWARD. so, you have to `flip` before `tokenizer.decode()`
|
79 |
+
해당 모델은 inference시 역순으로 예측합니다. (밥을 6시에 먹었어 -> 어 먹었 시에 여섯 을 밥) 때문에 `tokenizer.decode`를 수행하기 전에, `flip`으로 역순으로 치환해주세요.
|
80 |
+
|
81 |
+
Want see more detail follow this URL [KoGPT_num_converter](https://github.com/ddobokki/KoGPT_num_converter) and see `bart_inference.py` and `bart_train.py`
|
82 |
+
```python
|
83 |
+
class BartText2TextGenerationPipeline(Text2TextGenerationPipeline):
|
84 |
+
def postprocess(self, model_outputs, return_type=ReturnType.TEXT, clean_up_tokenization_spaces=False):
|
85 |
+
records = []
|
86 |
+
reversed_model_outputs = torch.flip(model_outputs["output_ids"][0], dims=[-1])
|
87 |
+
for output_ids in reversed_model_outputs:
|
88 |
+
if return_type == ReturnType.TENSORS:
|
89 |
+
record = {f"{self.return_name}_token_ids": output_ids}
|
90 |
+
elif return_type == ReturnType.TEXT:
|
91 |
+
record = {
|
92 |
+
f"{self.return_name}_text": self.tokenizer.decode(
|
93 |
+
output_ids,
|
94 |
+
skip_special_tokens=True,
|
95 |
+
clean_up_tokenization_spaces=clean_up_tokenization_spaces,
|
96 |
+
)
|
97 |
+
}
|
98 |
+
records.append(record)
|
99 |
+
return records
|
100 |
+
```
|
101 |
+
## Evaluation
|
102 |
+
Just using `evaluate-metric/bleu` and `evaluate-metric/rouge` in huggingface `evaluate` library
|
103 |
+
## How to Get Started With the Model
|
104 |
+
```python
|
105 |
+
from transformers.pipelines import Text2TextGenerationPipeline
|
106 |
+
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
|
107 |
+
texts = ["그러게 누가 6시까지 술을 마시래?"]
|
108 |
+
tokenizer = AutoTokenizer.from_pretrained(
|
109 |
+
inference_args.model_name_or_path,
|
110 |
+
)
|
111 |
+
model = AutoModelForSeq2SeqLM.from_pretrained(
|
112 |
+
inference_args.model_name_or_path,
|
113 |
+
)
|
114 |
+
# BartText2TextGenerationPipeline is implemented above (see 'Use')
|
115 |
+
seq2seqlm_pipeline = BartText2TextGenerationPipeline(model=model, tokenizer=tokenizer)
|
116 |
+
kwargs = {
|
117 |
+
"min_length": args.min_length,
|
118 |
+
"max_length": args.max_length,
|
119 |
+
"num_beams": args.beam_width,
|
120 |
+
"do_sample": args.do_sample,
|
121 |
+
"num_beam_groups": args.num_beam_groups,
|
122 |
+
}
|
123 |
+
pred = seq2seqlm_pipeline(texts, **kwargs)
|
124 |
+
print(pred)
|
125 |
+
# 그러게 누가 여섯 시까지 술을 마시래?
|
126 |
+
```
|