Youme21/Manhwa-NLLB-EnAr-v1

🥋 Manhwa-NLLB-EnAr (v1)

This is a fine-tuned translation model based on facebook/nllb-200-distilled-600M.
It is specifically trained to translate Manhwa, Manga, and Webnovels from English to Arabic.

Unlike general-purpose translators (Google/DeepL), this model understands:

• Murim/Cultivation Terms: Qi Deviation, Dantian, Young Master, Sect Leader.
• System/Leveling Terms: [System], Quests, Hunter Ranks (S-Rank, F-Rank).
• Slang & Tone: Handles aggressive dialogue ("Trash", "Bastard") and noble speech correctly.

📊 Performance

• BLEU Score: ~42.8 (on held-out Manhwa dataset)
• Training Data: ~80k high-quality manually cleaned sentence pairs.

🚀 How to Use

from transformers import AutoModelForSeq2SeqLM, AutoTokenizer

model_name = "Youme21/Manhwa-NLLB-EnAr-v1"

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

text = " I must circulate my mana to heal my internal injuries."

inputs = tokenizer(text, return_tensors="pt", padding=True)

# Translate to Arabic (arb_Arab)
translated_tokens = model.generate(
    **inputs,
    forced_bos_token_id=tokenizer.convert_tokens_to_ids("arb_Arab"),
    max_length=128
)

result = tokenizer.batch_decode(translated_tokens, skip_special_tokens=True)[0]
print(result)