Spaces:
Paused
Paused
| import gradio as gr | |
| import json | |
| from transformers import AutoModelForSeq2SeqLM, AutoTokenizer | |
| model_path = "anzorq/m2m100_418M_ft_ru-kbd_44K" | |
| src_lang="ru" | |
| tgt_lang="zu" | |
| tokenizer = AutoTokenizer.from_pretrained(model_path, src_lang=src_lang) | |
| model = AutoModelForSeq2SeqLM.from_pretrained(model_path) | |
| def translate(text, num_beams=4, num_return_sequences=4): | |
| inputs = tokenizer(text, return_tensors="pt") | |
| num_return_sequences = min(num_return_sequences, num_beams) | |
| translated_tokens = model.generate( | |
| **inputs, forced_bos_token_id=tokenizer.lang_code_to_id[tgt_lang], num_beams=num_beams, num_return_sequences=num_return_sequences | |
| ) | |
| translations = [] | |
| for translation in tokenizer.batch_decode(translated_tokens, skip_special_tokens=True): | |
| translations.append(translation) | |
| result = {"input":text, "translations":translations} | |
| return json.dumps(result) | |
| output = gr.Textbox() | |
| # with gr.Accordion("Advanced Options"): | |
| num_beams = gr.inputs.Slider(2, 10, step=1, label="Number of beams", default=4) | |
| num_return_sequences = gr.inputs.Slider(2, 10, step=1, label="Number of returned sentences", default=4) | |
| title = "Russian-Circassian translator demo" | |
| article = "<p style='text-align: center'>Want to help? Join the <a href='https://discord.gg/cXwv495r' target='_blank'>Discord server</a></p>" | |
| examples = [ | |
| ["Мы идем домой"], | |
| ["Сегодня хорошая погода"], | |
| ["Дети играют во дворе"], | |
| ["We live in a big house"], | |
| ["Tu es une bonne personne."], | |
| ["أين تعيش؟"], | |
| ["Bir şeyler yapmak istiyorum."], | |
| ["– Если я его отпущу, то ты вовек не сможешь его поймать, – заявил Сосруко."], | |
| ["Как только старик ушел, Сатаней пошла к Саусырыко."], | |
| ["我永远不会放弃你。"], | |
| ["우리는 소치에 살고 있습니다."], | |
| ] | |
| gr.Interface( | |
| fn=translate, | |
| inputs=["text", num_beams, num_return_sequences], | |
| outputs=output, | |
| title=title, | |
| examples=examples, | |
| article=article).launch() | |
| # import gradio as gr | |
| # title = "Русско-черкесский переводчик" | |
| # description = "Demo of a Russian-Circassian (Kabardian dialect) translator. <br>It is based on Facebook's <a href=\"https://about.fb.com/news/2020/10/first-multilingual-machine-translation-model/\">M2M-100 model</a> machine learning model, and has been trained on 45,000 Russian-Circassian sentence pairs. <br>It can also translate from 100 other languages to Circassian (English, French, Spanish, etc.), but less accurately. <br>The data corpus is constantly being expanded, and we need help in finding sentence sources, OCR, data cleaning, etc. <br>If you are interested in helping out with this project, please contact me at the link below.<br><br>This is only a demo, not a finished product. Translation quality is still low and will improve with time and more data.<br>45,000 sentence pairs is not enough to create an accurate machine translation model, and more data is needed.<br>You can help by finding sentence sources (books, web pages, etc.), scanning books, OCRing documents, data cleaning, and other tasks.<br><br>If you are interested in helping out with this project, contact me at the link below." | |
| # article = """<p style='text-align: center'><a href='https://arxiv.org/abs/1806.00187'>Scaling Neural Machine Translation</a> | <a href='https://github.com/pytorch/fairseq/'>Github Repo</a></p>""" | |
| # examples = [ | |
| # ["Мы идем домой"], | |
| # ["Сегодня хорошая погода"], | |
| # ["Дети играют во дворе"], | |
| # ["We live in a big house"], | |
| # ["Tu es une bonne personne."], | |
| # ["أين تعيش؟"], | |
| # ["Bir şeyler yapmak istiyorum."], | |
| # ] | |
| # gr.Interface.load("models/anzorq/m2m100_418M_ft_ru-kbd_44K", title=title, description=description, article=article, examples=examples).launch() |