M2M100 418M For GameTranslation Finetuned Zh En

🚀 M2M100-418M for Game Translation (Finetuned Zh-En)

This project is a finetuned version of Facebook/M2M100, aiming to provide better translation results for Chinese-to-English game translations. It's born from the efforts of the Amateur Modding Avenue community and benefits from the contributions of the Path of Wuxia modding team.

🚀 Quick Start

Prerequisites

• Python environment
• Install necessary libraries: transformers, torch

Sample Generation Script

from transformers import AutoModelForSeq2SeqLM, M2M100Tokenizer
import torch
device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
tokenizer = transformers.M2M100Tokenizer.from_pretrained("CadenzaBaron/M2M100-418M-for-GameTranslation-Finetuned-Zh-En")
model = AutoModelForSeq2SeqLM.from_pretrained("CadenzaBaron/M2M100-418M-for-GameTranslation-Finetuned-Zh-En")
model.to(device)
tokenizer.src_lang = "zh"
tokenizer.tgt_lang = "en"
test_string = "地阶上品遁术，施展后便可立于所持之剑上，以极快的速度自由飞行。"

inputs = tokenizer(test_string, return_tensors="pt").to(device)
translated_tokens = model.generate(**inputs, num_beams=10, do_sample=True)
translation = tokenizer.batch_decode(translated_tokens, skip_special_tokens=True)[0]

print("CH : ", test_string , " // EN : ", translation)

✨ Features

1. Tag Preservation: It should not mess up tags during the translation process.
2. Vocabulary for Cultivation and Martial Arts: It has basic cultivation/martial arts vocabulary, which is suitable for game translations in related themes.
3. Free of Charge: Nothing is locked behind a paywall.

📚 Documentation

Model Information

Inference Parameters

• src_lang: "zh"
• tgt_lang: "en"

Translation Sample and Comparison

You can check the translation sample and comparison with Google Translate and DeepL here.

📄 License

This project is licensed under the CC BY-NC 4.0 license.

⚠️ Important Note

• The model version has been bumped to V2, providing seemingly better translations. The legacy model is still available on the 'V1' branch.
• Before using the model, lower your expectations as it's not perfect. It should be sitting somewhere between Google Translate and DeepL.
• Considering the dataset is built from the work of modding groups (AMA and PoW Translation Team), who may not want their work to be reused for further AI training, the dataset will not be made public nor shared.