lmg-anon/vntl-7b-v0.3.1-lora

This is an experimental LLaMA2 7B lora created using the VNTL-v2.5-1k dataset.

This is a update of version 0.3:

• adamw_bnb_8bit -> adamw_8bit (this is the default in unsloth)
• 2 epoches -> 1 epoch (2 epoches seemed to increase eval loss)
• Added EOS after each translation pair.

Eval Loss: 0.72

This is an prompt example:

<<START>>
Name: Uryuu Shingo (瓜生 新吾) | Gender: Male | Aliases: Onii-chan (お兄ちゃん)
Name: Uryuu Sakuno (瓜生 桜乃) | Gender: Female
<<JAPANESE>>
[桜乃]: 『……ごめん』
<<ENGLISH>> (fidelity = absolute)
[Sakuno]: 『... Sorry.』</s>
<<JAPANESE>>
[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」
<<ENGLISH>> (fidelity = high)

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

[Shingo]: 「No, don't apologize. I'm just glad you're safe. You're so cute, Sakuno, I was worried sick.」