Bingsu's picture
Update README.md
57be1b1
|
raw
history blame
5.3 kB
---
pipeline_tag: sentence-similarity
tags:
- sentence-transformers
- feature-extraction
- sentence-similarity
- transformers
language:
- ko
license:
- mit
widget:
source_sentence: "λŒ€ν•œλ―Όκ΅­μ˜ μˆ˜λ„λŠ” μ„œμšΈμž…λ‹ˆλ‹€."
sentences:
- "미ꡭ의 μˆ˜λ„λŠ” λ‰΄μš•μ΄ μ•„λ‹™λ‹ˆλ‹€."
- "λŒ€ν•œλ―Όκ΅­μ˜ μˆ˜λ„ μš”κΈˆμ€ μ €λ ΄ν•œ νŽΈμž…λ‹ˆλ‹€."
- "μ„œμšΈμ€ λŒ€ν•œλ―Όκ΅­μ˜ μˆ˜λ„μž…λ‹ˆλ‹€."
---
# smartmind/roberta-ko-small-tsdae
This is a [sentence-transformers](https://www.SBERT.net) model: It maps sentences & paragraphs to a 256 dimensional dense vector space and can be used for tasks like clustering or semantic search.
Korean roberta small model pretrained with [TSDAE](https://arxiv.org/abs/2104.06979).
[TSDAE](https://arxiv.org/abs/2104.06979)둜 μ‚¬μ „ν•™μŠ΅λœ ν•œκ΅­μ–΄ robertaλͺ¨λΈμž…λ‹ˆλ‹€. λͺ¨λΈμ˜ κ΅¬μ‘°λŠ” [lassl/roberta-ko-small](https://huggingface.co/lassl/roberta-ko-small)κ³Ό λ™μΌν•©λ‹ˆλ‹€. ν† ν¬λ‚˜μ΄μ €λŠ” λ‹€λ¦…λ‹ˆλ‹€.
sentence-similarityλ₯Ό κ΅¬ν•˜λŠ” μš©λ„λ‘œ λ°”λ‘œ μ‚¬μš©ν•  μˆ˜λ„ 있고, λͺ©μ μ— 맞게 νŒŒμΈνŠœλ‹ν•˜μ—¬ μ‚¬μš©ν•  μˆ˜λ„ μžˆμŠ΅λ‹ˆλ‹€.
## Usage (Sentence-Transformers)
[sentence-transformers](https://www.SBERT.net)λ₯Ό μ„€μΉ˜ν•œ λ’€, λͺ¨λΈμ„ λ°”λ‘œ 뢈러올 수 μžˆμŠ΅λ‹ˆλ‹€.
```
pip install -U sentence-transformers
```
이후 λ‹€μŒμ²˜λŸΌ λͺ¨λΈμ„ μ‚¬μš©ν•  수 μžˆμŠ΅λ‹ˆλ‹€.
```python
from sentence_transformers import SentenceTransformer
sentences = ["This is an example sentence", "Each sentence is converted"]
model = SentenceTransformer('smartmind/roberta-ko-small-tsdae')
embeddings = model.encode(sentences)
print(embeddings)
```
λ‹€μŒμ€ sentence-transformers의 κΈ°λŠ₯을 μ‚¬μš©ν•˜μ—¬ μ—¬λŸ¬ λ¬Έμž₯의 μœ μ‚¬λ„λ₯Ό κ΅¬ν•˜λŠ” μ˜ˆμ‹œμž…λ‹ˆλ‹€.
```python
from sentence_transformers import util
sentences = [
"λŒ€ν•œλ―Όκ΅­μ˜ μˆ˜λ„λŠ” μ„œμšΈμž…λ‹ˆλ‹€.",
"미ꡭ의 μˆ˜λ„λŠ” λ‰΄μš•μ΄ μ•„λ‹™λ‹ˆλ‹€.",
"λŒ€ν•œλ―Όκ΅­μ˜ μˆ˜λ„ μš”κΈˆμ€ μ €λ ΄ν•œ νŽΈμž…λ‹ˆλ‹€.",
"μ„œμšΈμ€ λŒ€ν•œλ―Όκ΅­μ˜ μˆ˜λ„μž…λ‹ˆλ‹€.",
"였늘 μ„œμšΈμ€ ν•˜λ£¨μ’…μΌ λ§‘μŒ",
]
paraphrase = util.paraphrase_mining(model, sentences)
for score, i, j in paraphrase:
print(f"{sentences[i]}\t\t{sentences[j]}\t\t{score:.4f}")
```
```
λŒ€ν•œλ―Όκ΅­μ˜ μˆ˜λ„λŠ” μ„œμšΈμž…λ‹ˆλ‹€. μ„œμšΈμ€ λŒ€ν•œλ―Όκ΅­μ˜ μˆ˜λ„μž…λ‹ˆλ‹€. 0.7616
λŒ€ν•œλ―Όκ΅­μ˜ μˆ˜λ„λŠ” μ„œμšΈμž…λ‹ˆλ‹€. 미ꡭ의 μˆ˜λ„λŠ” λ‰΄μš•μ΄ μ•„λ‹™λ‹ˆλ‹€. 0.7031
λŒ€ν•œλ―Όκ΅­μ˜ μˆ˜λ„λŠ” μ„œμšΈμž…λ‹ˆλ‹€. λŒ€ν•œλ―Όκ΅­μ˜ μˆ˜λ„ μš”κΈˆμ€ μ €λ ΄ν•œ νŽΈμž…λ‹ˆλ‹€. 0.6594
미ꡭ의 μˆ˜λ„λŠ” λ‰΄μš•μ΄ μ•„λ‹™λ‹ˆλ‹€. μ„œμšΈμ€ λŒ€ν•œλ―Όκ΅­μ˜ μˆ˜λ„μž…λ‹ˆλ‹€. 0.6445
λŒ€ν•œλ―Όκ΅­μ˜ μˆ˜λ„ μš”κΈˆμ€ μ €λ ΄ν•œ νŽΈμž…λ‹ˆλ‹€. μ„œμšΈμ€ λŒ€ν•œλ―Όκ΅­μ˜ μˆ˜λ„μž…λ‹ˆλ‹€. 0.4915
미ꡭ의 μˆ˜λ„λŠ” λ‰΄μš•μ΄ μ•„λ‹™λ‹ˆλ‹€. λŒ€ν•œλ―Όκ΅­μ˜ μˆ˜λ„ μš”κΈˆμ€ μ €λ ΄ν•œ νŽΈμž…λ‹ˆλ‹€. 0.4785
μ„œμšΈμ€ λŒ€ν•œλ―Όκ΅­μ˜ μˆ˜λ„μž…λ‹ˆλ‹€. 였늘 μ„œμšΈμ€ ν•˜λ£¨μ’…μΌ λ§‘μŒ 0.4119
λŒ€ν•œλ―Όκ΅­μ˜ μˆ˜λ„λŠ” μ„œμšΈμž…λ‹ˆλ‹€. 였늘 μ„œμšΈμ€ ν•˜λ£¨μ’…μΌ λ§‘μŒ 0.3520
미ꡭ의 μˆ˜λ„λŠ” λ‰΄μš•μ΄ μ•„λ‹™λ‹ˆλ‹€. 였늘 μ„œμšΈμ€ ν•˜λ£¨μ’…μΌ λ§‘μŒ 0.2550
λŒ€ν•œλ―Όκ΅­μ˜ μˆ˜λ„ μš”κΈˆμ€ μ €λ ΄ν•œ νŽΈμž…λ‹ˆλ‹€. 였늘 μ„œμšΈμ€ ν•˜λ£¨μ’…μΌ λ§‘μŒ 0.1896
```
## Usage (HuggingFace Transformers)
[sentence-transformers](https://www.SBERT.net)λ₯Ό μ„€μΉ˜ν•˜μ§€ μ•Šμ€ μƒνƒœλ‘œλŠ” λ‹€μŒμ²˜λŸΌ μ‚¬μš©ν•  수 μžˆμŠ΅λ‹ˆλ‹€.
```python
from transformers import AutoTokenizer, AutoModel
import torch
def cls_pooling(model_output, attention_mask):
return model_output[0][:,0]
# Sentences we want sentence embeddings for
sentences = ['This is an example sentence', 'Each sentence is converted']
# Load model from HuggingFace Hub
tokenizer = AutoTokenizer.from_pretrained('smartmind/roberta-ko-small-tsdae')
model = AutoModel.from_pretrained('smartmind/roberta-ko-small-tsdae')
# Tokenize sentences
encoded_input = tokenizer(sentences, padding=True, truncation=True, return_tensors='pt')
# Compute token embeddings
with torch.no_grad():
model_output = model(**encoded_input)
# Perform pooling. In this case, cls pooling.
sentence_embeddings = cls_pooling(model_output, encoded_input['attention_mask'])
print("Sentence embeddings:")
print(sentence_embeddings)
```
## Evaluation Results
[klue](https://huggingface.co/datasets/klue) STS 데이터에 λŒ€ν•΄ λ‹€μŒ 점수λ₯Ό μ–»μ—ˆμŠ΅λ‹ˆλ‹€. 이 데이터에 λŒ€ν•΄ νŒŒμΈνŠœλ‹ν•˜μ§€ **μ•Šμ€** μƒνƒœλ‘œ κ΅¬ν•œ μ μˆ˜μž…λ‹ˆλ‹€.
|split|cosine_pearson|cosine_spearman|euclidean_pearson|euclidean_spearman|manhattan_pearson|manhattan_spearman|dot_pearson|dot_spearman|
|-----|--------------|---------------|-----------------|------------------|-----------------|------------------|-----------|------------|
|train|0.8735|0.8676|0.8268|0.8357|0.8248|0.8336|0.8449|0.8383|
|validation|0.5409|0.5349|0.4786|0.4657|0.4775|0.4625|0.5284|0.5252|
## Full Model Architecture
```
SentenceTransformer(
(0): Transformer({'max_seq_length': 508, 'do_lower_case': False}) with Transformer model: RobertaModel
(1): Pooling({'word_embedding_dimension': 256, 'pooling_mode_cls_token': True, 'pooling_mode_mean_tokens': False, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False})
)
```
## Citing & Authors
<!--- Describe where people can find more information -->