juierror's picture
Update README.md
2f00d2e
|
raw
history blame
1.6 kB
metadata
language: en
datasets:
  - wikisql
widget:
  - text: 'question: get people name with age equal 25 table: id, name, age'
license: apache-2.0

How to use

from typing import List
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

tokenizer = AutoTokenizer.from_pretrained("juierror/flan-t5-text2sql-with-schema")
model = AutoModelForSeq2SeqLM.from_pretrained("juierror/flan-t5-text2sql-with-schema")

def prepare_input(question: str, table: List[str]):
    table_prefix = "table:"
    question_prefix = "question:"
    join_table = ",".join(table)
    inputs = f"{question_prefix} {question} {table_prefix} {join_table}"
    input_ids = tokenizer(inputs, max_length=512, return_tensors="pt").input_ids
    return input_ids

def inference(question: str, table: List[str]) -> str:
    input_data = prepare_input(question=question, table=table)
    input_data = input_data.to(model.device)
    outputs = model.generate(inputs=input_data, num_beams=10, top_k=10, max_length=700)
    result = tokenizer.decode(token_ids=outputs[0], skip_special_tokens=True)
    return result

print(inference(question="get people name with age equal 25", table=["id", "name", "age"]))

PS. From this discussion, I think the base model that I use for finetune did not support the token <, so this might not be a good model to do this tasks.

However, you might consider to use work around method from vonjack.