Vntl Llama3 8b V2 Hf

🚀 LLaMA 3 Youko QLoRA Fine-tune for Japanese Visual Novel Translation

This project is a QLoRA fine-tune of LLaMA 3 Youko, leveraging a new version of the VNTL dataset. Its core objective is to enhance the performance of large language models (LLMs) in translating Japanese visual novels into English.

🚀 Quick Start

This fine-tuned model is designed to improve the translation of Japanese visual novels to English. It uses the LLaMA 3 prompt format and offers better accuracy and stability compared to the previous version.

✨ Features

• Enhanced Dataset: The new version of the VNTL dataset has been rebuilt and expanded from the ground up, leading to better performance in terms of accuracy and stability.
• Default Prompt Format: Switched to the default LLaMA 3 prompt format, which resolves issues users had with the custom one.
• Multi - line Translation Support: Added proper support for multi - line translations, while the old version only handled single lines.
• Improved Accuracy: Overall better translation accuracy, although the translations tend to be more literal.

📦 Installation

No installation steps are provided in the original document, so this section is skipped.

💻 Usage Examples

Basic Usage

This fine - tune uses the LLaMA 3 prompt format. Here is an example for translation:

<|begin_of_text|><|start_header_id|>Metadata<|end_header_id|>

[character] Name: Uryuu Shingo (瓜生 新吾) | Gender: Male | Aliases: Onii-chan (お兄ちゃん)
[character] Name: Uryuu Sakuno (瓜生 桜乃) | Gender: Female<|eot_id|><|start_header_id|>Japanese<|end_header_id|>

[桜乃]: 『……ごめん』<|eot_id|><|start_header_id|>English<|end_header_id|>

[Sakuno]: 『... Sorry.』<|eot_id|><|start_header_id|>Japanese<|end_header_id|>

[新吾]: 「ううん、こう言っちゃなんだけど、迷子でよかったよ。桜乃は可愛いから、いろいろ心配しちゃってたんだぞ俺」<|eot_id|><|start_header_id|>English<|end_header_id|>

[Shingo]: "Nah, I know it’s weird to say this, but I’m glad you got lost. You’re so cute, Sakuno, so I was really worried about you."<|eot_id|>

Advanced Usage

The Metadata section isn't limited to character information. You can also add trivia and teach the model the correct way to pronounce words it struggles with. Here's an example:

<|begin_of_text|><|start_header_id|>Metadata<|end_header_id|>

[character] Name: Uryuu Shingo (瓜生 新吾) | Gender: Male | Aliases: Onii-chan (お兄ちゃん)
[character] Name: Uryuu Sakuno (瓜生 桜乃) | Gender: Female
[element] Name: Murasamemaru (叢雨丸) | Type: Quality<|eot_id|><|start_header_id|>Japanese<|end_header_id|>

[桜乃]: 『……ごめん』<|eot_id|><|start_header_id|>English<|end_header_id|>

[Sakuno]: 『... Sorry.』<|eot_id|><|start_header_id|>Japanese<|end_header_id|>

[新吾]: 「ううん、こう言っちゃなんだけど、迷子でよかったよ。桜乃は叢雨丸いから、いろいろ心配しちゃってたんだぞ俺」<|eot_id|><|start_header_id|>English<|end_header_id|>

The generated translation for that prompt, with temperature 0, is:

[Shingo]: "Nah, I know it’s not the best thing to say, but I’m glad you got lost. Sakuno’s Murasamemaru, so I was really worried about you, you know?"

📚 Documentation

Sampling Recommendations

For optimal results, it's highly recommended to use neutral sampling parameters (temperature 0 with no repetition penalty) when using this model.

Notes

This new version of VNTL 8B has been rebuilt and expanded. It outperforms the previous version in accuracy and stability, making far fewer mistakes even at high temperatures (though temperature 0 is still recommended for the best accuracy). The translations are more accurate but tend to be more literal compared to the previous version.

🔧 Technical Details

This fine - tune was done using similar hyperparameters as the previous version. The only difference is the dataset, which is a brand - new one.

📄 License

The license for this project is llama3.

⚠️ Important Note

While the translations are more accurate, they tend to be more literal compared to the previous version.

💡 Usage Tip

For optimal results, it's highly recommended to use neutral sampling parameters (temperature 0 with no repetition penalty) when using this model.