app.py

import spaces
import gradio as gr
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
from flores import code_mapping
import platform
import torch

device = "cpu" if platform.system() == "Darwin" else "cuda"
MODEL_NAME = "facebook/nllb-200-3.3B"

code_mapping = dict(sorted(code_mapping.items(), key=lambda item: item[1]))
flores_codes = list(code_mapping.keys())


def load_model():
    model = AutoModelForSeq2SeqLM.from_pretrained(MODEL_NAME).to(device)
    return model


model = load_model()


def load_tokenizer(src_lang, tgt_lang):
    tokenizer = AutoTokenizer.from_pretrained(
        MODEL_NAME, src_lang=code_mapping[src_lang], tgt_lang=code_mapping[tgt_lang]
    )
    return tokenizer


@spaces.GPU
def translate(
    text: str,
    src_lang: str,
    tgt_lang: str,
    window_size: int = 800,
    overlap_size: int = 200,
):
    tokenizer = load_tokenizer(src_lang, tgt_lang)

    input_tokens = (
        tokenizer(text, return_tensors="pt").input_ids[0].cpu().numpy().tolist()
    )
    translated_chunks = []

    for i in range(0, len(input_tokens), window_size - overlap_size):
        window = input_tokens[i : i + window_size]
        translated_chunk = model.generate(
            input_ids=torch.tensor([window]).to(device),
            forced_bos_token_id=tokenizer.lang_code_to_id[code_mapping[tgt_lang]],
            max_length=window_size,
            num_return_sequences=1,
        )
        translated_chunk = tokenizer.decode(
            translated_chunk[0], skip_special_tokens=True
        )
        translated_chunks.append(translated_chunk)

    return " ".join(translated_chunks)


description = """
No Language Left Behind (NLLB) is a series of open-source models aiming to provide high-quality translations between 200 languages.
This demo application allows you to use the NLLB model to translate text between a source and target language.

## Notes 

- Whilst the model supports 200 languages, the quality of translations may vary between languages. 
- "Low Resource" languages (languages which are less present on the internet and have a lower amount of investment) may have lower quality translations.
- The demo uses a sliding window approach to handle longer texts.
"""

instructions = """
1. Select the source and target language from the dropdown menus.
2. Enter the text you would like to translate.
3. Click the 'Translate text' button.
"""
with gr.Blocks() as demo:
    gr.Markdown("# No Language Left Behind (NLLB) Translation Demo")
    gr.Markdown(description)
    gr.Markdown("## Instructions")
    gr.Markdown(instructions)
    with gr.Row():
        src_lang = gr.Dropdown(label="Source Language", choices=flores_codes)
        target_lang = gr.Dropdown(label="Target Language", choices=flores_codes)
    with gr.Row():
        input_text = gr.Textbox(label="Input Text", lines=6)
    with gr.Row():
        btn = gr.Button("Translate text")
    with gr.Row():
        output = gr.Textbox(label="Output Text", lines=6)
    btn.click(
        translate,
        inputs=[input_text, src_lang, target_lang],
        outputs=output,
    )
demo.launch()