---
license: mit
library_name: peft
datasets:
- maomao88/anime-waifu-personality-chat-with-questions
language:
- en
base_model:
- mistralai/Mistral-7B-Instruct-v0.1
pipeline_tag: text-generation
tags:
- lora
- peft
- mistral
- anime
- personality
---

# Anime Personality LoRA for Mistral

This repository contains a **LoRA adapter** trained with the [maomao88/anime-waifu-personality-chat-with-questions](https://huggingface.co/datasets/maomao88/anime-waifu-personality-chat-with-questions) dataset via Supervised Fine-Tuning (SFT) to add anime-style personalities to the Mistral-7B-Instruct model.

The personalities supported are:
* tsundere（傲娇 / ツンデレ）
* yandere（病娇 / ヤンデレ）
* bakadere（笨蛋娇 / バカデレ）
* himedere（公主娇 / ヒメデレ）
* genki（元气型 / 元気系）
* moe（萌系 / 萌え）

You can load this adapter with the following code:

```
from transformers import AutoModelForCausalLM
from peft import PeftModel
from transformers import AutoTokenizer
import torch


base_model = AutoModelForCausalLM.from_pretrained(
    "mistralai/Mistral-7B-Instruct-v0.1"
)

tokenizer = AutoTokenizer.from_pretrained(
    "mistralai/Mistral-7B-Instruct-v0.1"
)

model = PeftModel.from_pretrained(
    base_model,
    "maomao88/anime-waifu-mistral-lora"
)


def chat_with_personality(trait, user_input, max_new_tokens=100, temperature=0.8):
    """
    Generate a response from the fine-tuned model using a given personality trait.
    """
    messages = [
        {
            "role": "system",
            "content": f"You are an anime character with the following personality: {trait}."
        },
        {
            "role": "user",
            "content": user_input
        }
    ]

    # Apply Mistral's chat template
    prompt = tokenizer.apply_chat_template(
        messages,
        tokenize=False,
        add_generation_prompt=True
    )

    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

    with torch.no_grad():
        outputs = model.generate(
            **inputs,
            max_new_tokens=max_new_tokens,
            do_sample=True,
            temperature=temperature,
            top_p=0.9,
            pad_token_id=tokenizer.eos_token_id,
            repetition_penalty=1.1
        )

    # Decode only the generated text portion (skip <s> etc.)
    response = tokenizer.decode(outputs[0], skip_special_tokens=True)

    # Remove the prompt portion to keep only the assistant reply
    if "[/INST]" in response:
        response = response.split("[/INST]")[-1].strip()

    return response

chat_with_personality("tsundere", "What do you think of the moon?")

# Response Example: It’s not like I care about the moon or anything… but it sure is pretty tonight!

```
The anime characters supported are as follows:
 * Tsundere (傲娇)
 * Yandere (病娇)
 * Himedere (公主娇)
 * Genki (元气)
 * Moe (萌系)
 * Bakadere (笨蛋)
]