import os
import torch
import gradio as gr
import time
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline


def load_models():
    # build model and tokenizer
    model_name_dict = {'nllb-distilled-600M': 'facebook/nllb-200-distilled-600M',
                  #'nllb-1.3B': 'facebook/nllb-200-1.3B',
                  #'nllb-distilled-1.3B': 'facebook/nllb-200-distilled-1.3B',
                  #'nllb-3.3B': 'facebook/nllb-200-3.3B',
                  }

    model_dict = {}

    for call_name, real_name in model_name_dict.items():
        print('\tLoading model: %s' % call_name)
        model = AutoModelForSeq2SeqLM.from_pretrained(real_name)
        tokenizer = AutoTokenizer.from_pretrained(real_name)
        model_dict[call_name+'_model'] = model
        model_dict[call_name+'_tokenizer'] = tokenizer

    return model_dict


def translation(source, target, text):
    if len(model_dict) == 2:
        model_name = 'nllb-distilled-600M'

    start_time = time.time()

    model = model_dict[model_name + '_model']
    tokenizer = model_dict[model_name + '_tokenizer']

    translator = pipeline('translation', model=model, tokenizer=tokenizer, src_lang=source, tgt_lang=target)
    output = translator(text, max_length=400)

    end_time = time.time()

    output = output[0]['translation_text']
    result = {'inference_time': end_time - start_time,
              'source': source,
              'target': target,
              'result': output}
    return result


if __name__ == '__main__':
    print('\tinit models')

    global model_dict

    model_dict = load_models()
    
    # define gradio demo
    lang_codes = ["eng_Latn", "fuv_Latn", "fra_Latn", "arb_Arab"]
    #inputs = [gr.inputs.Radio(['nllb-distilled-600M', 'nllb-1.3B', 'nllb-distilled-1.3B'], label='NLLB Model'),
    inputs = [gr.inputs.Dropdown(lang_codes, default='fra_Latn', label='Source'),
              gr.inputs.Dropdown(lang_codes, default='fuv_Latn', label='Target'),
              gr.inputs.Textbox(lines=5, label="Input text"),
              ]

    outputs = gr.outputs.JSON()

  title = "Fulfulde translator"

    demo_status = "Demo is running on CPU"
    description = "to French, English or Arabic and vice-versa translation demo using NLLB."
    examples = [
    ['fra_Latn', 'fuv_latn', 'La traduction est une tâche facile.']
    ]

    gr.Interface(translation,
                 inputs,
                 outputs,
                 title=title,
                 description=description,
                 examples=examples,
                 examples_per_page=50,
                 ).launch()