import gradio as gr

import torch
import transformers



# saved_model
def load_model(model_path):
    saved_data = torch.load(
        model_path,
        map_location="cpu"
    )

    bart_best = saved_data["model"]
    train_config = saved_data["config"]
    tokenizer = transformers.PreTrainedTokenizerFast.from_pretrained('gogamza/kobart-base-v1')

    ## Load weights.
    model = transformers.BartForConditionalGeneration.from_pretrained('gogamza/kobart-base-v1')
    model.load_state_dict(bart_best)

    return model, tokenizer


# main
def inference(prompt):
    model_path = "./kobart-model-logical.pth"

    model, tokenizer = load_model(
        model_path=model_path
        )

    input_ids = tokenizer.encode(prompt)
    input_ids = torch.tensor(input_ids)
    input_ids = input_ids.unsqueeze(0)
    output = model.generate(input_ids)
    output = tokenizer.decode(output[0], skip_special_tokens=True)    

    return output


demo = gr.Interface(
    fn=inference, 
    inputs="text", 
    outputs="text" #return 값
    ).launch(share=True) # launch(share=True)를 설정하면 외부에서 접속 가능한 링크가 생성됨

demo.launch()