juierror's picture
Update README.md
177b4a5
---
language: en
widget:
- text: >-
convert question and table into SQL query. tables: people_name(id,name),
people_age(people_id,age). question: how many people with name jui and age
less than 25
license: cc-by-sa-4.0
pipeline_tag: text2text-generation
inference:
parameters:
max_length: 512
num_beams: 10
top_k: 10
---
This is an upgraded version of [https://huggingface.co/juierror/flan-t5-text2sql-with-schema](https://huggingface.co/juierror/flan-t5-text2sql-with-schema).
It supports the '<' sign and can handle multiple tables.
# How to use
```python
from typing import List
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
tokenizer = AutoTokenizer.from_pretrained("juierror/flan-t5-text2sql-with-schema-v2")
model = AutoModelForSeq2SeqLM.from_pretrained("juierror/flan-t5-text2sql-with-schema-v2")
def get_prompt(tables, question):
prompt = f"""convert question and table into SQL query. tables: {tables}. question: {question}"""
return prompt
def prepare_input(question: str, tables: Dict[str, List[str]]):
tables = [f"""{table_name}({",".join(tables[table_name])})""" for table_name in tables]
tables = ", ".join(tables)
prompt = get_prompt(tables, question)
input_ids = tokenizer(prompt, max_length=512, return_tensors="pt").input_ids
return input_ids
def inference(question: str, tables: Dict[str, List[str]]) -> str:
input_data = prepare_input(question=question, tables=tables)
input_data = input_data.to(model.device)
outputs = model.generate(inputs=input_data, num_beams=10, top_k=10, max_length=512)
result = tokenizer.decode(token_ids=outputs[0], skip_special_tokens=True)
return result
print(inference("how many people with name jui and age less than 25", {
"people_name": ["id", "name"],
"people_age": ["people_id", "age"]
}))
print(inference("what is id with name jui and age less than 25", {
"people_name": ["id", "name", "age"]
})))
```
# Dataset
- [CoSQL](https://yale-lily.github.io/cosql)
- [spider](https://yale-lily.github.io/spider)
- [SParC](https://yale-lily.github.io/sparc)