import gradio as gr
from transformers import pipeline

# pipeline = pipeline(task="image-classification", model="julien-c/hotdog-not-hotdog")

# def predict(input_img):
#     predictions = pipeline(input_img)
#     return input_img, {p["label"]: p["score"] for p in predictions} 

from typing import List
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

tokenizer = AutoTokenizer.from_pretrained("juierror/text-to-sql-with-table-schema")
model = AutoModelForSeq2SeqLM.from_pretrained("juierror/text-to-sql-with-table-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=700, return_tensors="pt").input_ids
    return input_ids

def inference(question: str, table: str) -> str:
    cols = table.split(",")
    input_data = prepare_input(question=question, table=cols)
    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"]))

gradio_app = gr.Interface(
    inference,
    inputs=["textbox", "textbox"],
    outputs="label",
    title="Text To SQL",
)

if __name__ == "__main__":
    gradio_app.launch()