File size: 4,991 Bytes
4efa655 a15a665 4efa655 2b5dfe5 4efa655 |
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 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 |
---
language:
- ko
license: apache-2.0
tags:
- text2sql
- spider
- korean
- llama
- text-generation
- table-question-answering
datasets:
- spider
- huggingface-KREW/spider-ko
base_model: unsloth/Meta-Llama-3.1-8B-Instruct
model-index:
- name: Llama-3.1-8B-Spider-SQL-Ko
results:
- task:
type: text2sql
name: Text to SQL
dataset:
name: Spider (Korean)
type: text2sql
metrics:
- type: exact_match
value: 42.65
- type: execution_accuracy
value: 65.47
---
# Llama-3.1-8B-Spider-SQL-Ko
ํ๊ตญ์ด ์ง๋ฌธ์ SQL ์ฟผ๋ฆฌ๋ก ๋ณํํ๋ Text-to-SQL ๋ชจ๋ธ์
๋๋ค.
[Spider](https://yale-lily.github.io/spider) train ๋ฐ์ดํฐ์
์ ํ๊ตญ์ด๋ก ๋ฒ์ญํ [spider-ko](https://huggingface.co/datasets/huggingface-KREW/spider-ko) ๋ฐ์ดํฐ์
์ ํ์ฉํ์ฌ ๋ฏธ์ธ์กฐ์ ํ์์ต๋๋ค.
## ๐ ์ฃผ์ ์ฑ๋ฅ
Spider ํ๊ตญ์ด ๊ฒ์ฆ ๋ฐ์ดํฐ์
(1,034๊ฐ) ํ๊ฐ ๊ฒฐ๊ณผ:
- **์ ํ ์ผ์น์จ**: 42.65% (441/1034)
- **์คํ ์ ํ๋**: 65.47% (677/1034)
> ๐ก ์คํ ์ ํ๋๊ฐ ์ ํ ์ผ์น์จ๋ณด๋ค ๋์ ์ด์ ๋, SQL ๋ฌธ๋ฒ์ด ๋ค๋ฅด๋๋ผ๋ ๋์ผํ ๊ฒฐ๊ณผ๋ฅผ ๋ฐํํ๋ ๊ฒฝ์ฐ๊ฐ ๋ง๊ธฐ ๋๋ฌธ์
๋๋ค.
## ๐ ๋ฐ๋ก ์์ํ๊ธฐ
```python
from unsloth import FastLanguageModel
# ๋ชจ๋ธ ๋ถ๋ฌ์ค๊ธฐ
model, tokenizer = FastLanguageModel.from_pretrained(
model_name="huggingface-KREW/Llama-3.1-8B-Spider-SQL-Ko",
max_seq_length=2048,
dtype=None,
load_in_4bit=True,
)
# ํ๊ตญ์ด ์ง๋ฌธ โ SQL ๋ณํ
question = "๊ฐ์๋ ๋ช ๋ช
์ด ์๋์?"
schema = """ํ
์ด๋ธ: singer
์ปฌ๋ผ: singer_id, name, country, age"""
prompt = f"""๋ฐ์ดํฐ๋ฒ ์ด์ค ์คํค๋ง:
{schema}
์ง๋ฌธ: {question}
SQL:"""
# ๊ฒฐ๊ณผ: SELECT count(*) FROM singer
```
## ๐ ๋ชจ๋ธ ์๊ฐ
- **๊ธฐ๋ฐ ๋ชจ๋ธ**: Llama 3.1 8B Instruct (4bit ์์ํ)
- **ํ์ต ๋ฐ์ดํฐ**: [spider-ko](https://huggingface.co/datasets/huggingface-KREW/spider-ko) (1-epoch)
- **์ง์ DB**: 166๊ฐ์ ๋ค์ํ ๋๋ฉ์ธ ๋ฐ์ดํฐ๋ฒ ์ด์ค ( [spider dataset](https://yale-lily.github.io/spider) )
- **ํ์ต ๋ฐฉ๋ฒ**: LoRA (r=16, alpha=32)
## ๐ฌ ํ์ฉ ์์
### ๊ธฐ๋ณธ ์ฌ์ฉ๋ฒ
```python
def generate_sql(question, schema_info):
"""ํ๊ตญ์ด ์ง๋ฌธ์ SQL๋ก ๋ณํ"""
prompt = f"""๋ค์ ๋ฐ์ดํฐ๋ฒ ์ด์ค ์คํค๋ง๋ฅผ ์ฐธ๊ณ ํ์ฌ ์ง๋ฌธ์ ๋ํ SQL ์ฟผ๋ฆฌ๋ฅผ ์์ฑํ์ธ์.
### ๋ฐ์ดํฐ๋ฒ ์ด์ค ์คํค๋ง:
{schema_info}
### ์ง๋ฌธ: {question}
### SQL ์ฟผ๋ฆฌ:"""
messages = [{"role": "user", "content": prompt}]
inputs = tokenizer.apply_chat_template(messages, tokenize=True, add_generation_prompt=True, return_tensors="pt")
outputs = model.generate(inputs, max_new_tokens=150, temperature=0.1)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
return response.split("### SQL ์ฟผ๋ฆฌ:")[-1].strip()
```
### ์ค์ ์ฌ์ฉ ์์
```python
# ์์ 1: ์ง๊ณ ํจ์
question = "๋ถ์์ฅ๋ค ์ค 56์ธ๋ณด๋ค ๋์ด๊ฐ ๋ง์ ์ฌ๋์ด ๋ช ๋ช
์
๋๊น?"
# ๊ฒฐ๊ณผ: SELECT count(*) FROM head WHERE age > 56
# ์์ 2: ์กฐ์ธ
question = "๊ฐ์ฅ ๋ง์ ๋ํ๋ฅผ ๊ฐ์ตํ ๋์์ ์ํ๋ ๋ฌด์์ธ๊ฐ์?"
# ๊ฒฐ๊ณผ: SELECT T1.Status FROM city AS T1 JOIN farm_competition AS T2 ON T1.City_ID = T2.Host_city_ID GROUP BY T2.Host_city_ID ORDER BY COUNT(*) DESC LIMIT 1
# ์์ 3: ์๋ธ์ฟผ๋ฆฌ
question = "๊ธฐ์
๊ฐ๊ฐ ์๋ ์ฌ๋๋ค์ ์ด๋ฆ์ ๋ฌด์์
๋๊น?"
# ๊ฒฐ๊ณผ: SELECT Name FROM people WHERE People_ID NOT IN (SELECT People_ID FROM entrepreneur)
```
## โ ๏ธ ์ฌ์ฉ ์ ์ฃผ์์ฌํญ
### ์ ํ์ฌํญ
- โ
์์ด ํ
์ด๋ธ/์ปฌ๋ผ๋ช
์ฌ์ฉ (ํ๊ตญ์ด ์ง๋ฌธ โ ์์ด SQL)
- โ
Spider ๋ฐ์ดํฐ์
๋๋ฉ์ธ์ ์ต์ ํ
- โ NoSQL, ๊ทธ๋ํ DB ๋ฏธ์ง์
- โ ๋งค์ฐ ๋ณต์กํ ์ค์ฒฉ ์ฟผ๋ฆฌ๋ ์ ํ๋ ํ๋ฝ
## ๐ง ๊ธฐ์ ์ฌ์
### ํ์ต ํ๊ฒฝ
- **GPU**: NVIDIA Tesla T4 (16GB)
- **ํ์ต ์๊ฐ**: ์ฝ 4์๊ฐ
- **๋ฉ๋ชจ๋ฆฌ ์ฌ์ฉ**: ์ต๋ 7.6GB VRAM
### ํ์ดํผํ๋ผ๋ฏธํฐ
```python
training_args = {
"per_device_train_batch_size": 2,
"gradient_accumulation_steps": 4,
"learning_rate": 5e-4,
"num_train_epochs": 1,
"optimizer": "adamw_8bit",
"lr_scheduler_type": "cosine",
"warmup_ratio": 0.05
}
lora_config = {
"r": 16,
"lora_alpha": 32,
"lora_dropout": 0,
"target_modules": ["q_proj", "k_proj", "v_proj", "o_proj",
"gate_proj", "up_proj", "down_proj"]
}
```
## ๐ ์ฐธ๊ณ ์๋ฃ
### ์ธ์ฉ
```bibtex
@misc{llama31_spider_sql_ko_2025,
title={Llama-3.1-8B-Spider-SQL-Ko: Korean Text-to-SQL Model},
author={[Sohyun Sim, Youngjun Cho, Seongwoo Choi]},
year={2025},
publisher={Hugging Face KREW},
url={https://huggingface.co/huggingface-KREW/Llama-3.1-8B-Spider-SQL-Ko}
}
```
### ๊ด๋ จ ๋
ผ๋ฌธ
- [Spider: A Large-Scale Human-Labeled Dataset](https://arxiv.org/abs/1809.08887) (Yu et al., 2018)
## ๐ค ๊ธฐ์ฌ์
[@sim-so](https://huggingface.co/sim-so), [@choincnp](https://huggingface.co/choincnp), [@nuatmochoi](https://huggingface.co/nuatmochoi)
|