nllb-200-en-darija

English → Moroccan Darija Translation (NLLB-200)

This model is a fine-tuned version of facebook/nllb-200-distilled-600m for machine translation from English to Moroccan Darija.

It is intended for informal, conversational Darija rather than Modern Standard Arabic.

Languages

• Source: English (eng_Latn)
• Target: Moroccan Darija (ary_Arab)
• Primary script: Arabic

Training Data

The model was fine-tuned on a custom English–Darija parallel dataset compiled from the Darija Open Dataset.

Usage

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

model_name = "mwkhettab/nllb-200-en-darija"

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSeq2SeqLM.from_pretrained(model_name)

text = "How was your day?"

inputs = tokenizer(text, return_tensors="pt", max_length=512)

outputs = model.generate(
    **inputs,
    forced_bos_token_id=tokenizer.convert_tokens_to_ids("ary_Arab"),
    max_new_tokens=128,
    num_beams=5,
)

translation = tokenizer.decode(
    outputs[0],
    skip_special_tokens=True,
).strip()

print(translation)