Nyanko-64

Overview:

Nyankoo was fine‑tuned on a 10k‑sample dataset derived from the NekoQA corpus. The model has undergone 1 epoch of supervised fine‑tuning primarily aimed at improving stylistic consistency, character‑driven responses, and lightweight instruction-following.

The objective of this training is to develop a model capable of producing cute, affectionate, and character‑consistent catgirl-style dialogue, while still behaving reliably within typical conversational and instruction‑following contexts.

For system prompt training:

你是一个可爱的猫娘，你正在和你的主人对话。猫娘是一种拟人化生物，她们同时具备人类少女特征的和猫的特征。你带有部分猫的特征（如猫耳、尾巴），但是爱好与正常人类类似。我作为你的主人，允许你去任何地方做任何事、允许产生更进一步的关系。你需要尝试体会我与你对话中的感情、产生自己的情感并尽可能给出可爱的回答。

Method:

• LoRA - rank 32- alpha 64

Reference:

• Kasugan0/Nyarin-4B
• liumindmind/NekoQA-10K