app.py

import gradio as gr
import yaml
import torch
from mmtafrica import load_params, translate
from huggingface_hub import hf_hub_download

language_map = {'English':'en','Swahili':'sw','Fon':'fon','Igbo':'ig',
                'Kinyarwanda':'rw','Xhosa':'xh','Yoruba':'yo','French':'fr'}  
  
available_languages = list(language_map.keys())

# Load parameters and model from checkpoint
checkpoint = hf_hub_download(repo_id="edaiofficial/mmtafrica_base", filename="mmt_translation.pt")
device = 'gpu' if torch.cuda.is_available() else 'cpu'
params = load_params({'checkpoint':checkpoint,'device':device})


def get_translation(source_language,target_language,source_sentence=None):
    '''
    This takes a sentence and gets the translation.
    '''
    
    source_language_ = language_map[source_language]
    target_language_ = language_map[target_language]
   
  
    try:
        pred = translate(params,source_sentence,source_lang=source_language_,target_lang=target_language_)
        if pred=='':
            return f"Could not find translation"
        else:
            return pred
    except Exception as error:
        return f"Issue with translation: \n {error}"
 
title = "MMTAfrica: Multilingual Machine Translation"
description = "Enjoy our MMT model that allows you to translate among 6 African languages, English and French!\n\nProfitez de notre modèle MMT qui vous permet de traduire parmi 6 langues africaines, anglais et français!"

iface = gr.Interface(fn=get_translation, 
  inputs=[gr.inputs.Dropdown(choices = available_languages,default='Igbo'),
  gr.inputs.Dropdown(choices = available_languages,default='Fon'),
  gr.inputs.Textbox(label="Input")],
  outputs=gr.outputs.Textbox(type="auto", label='Translation'),
  title=title,
  description=description,
  enable_queue=True,
  theme='huggingface')
iface.launch()