---
license: apache-2.0
language:
- en
datasets:
- Muizah/anime-bias-dataset
base_model:
- Qwen/Qwen2.5-3B-Instruct
tags:
- anime
- persona
- text-generayion
- qwen2.5
---
tags:
- qwen
- qwen2.5
- anime
- persona
- merged
- text-generation
- instruction-tuned
license: apache-2.0
language:
- en
base_model: Qwen/Qwen2.5-3B-Instruct
library_name: transformers
pipeline_tag: text-generation
---

#  Anime-Friend-Merged

**Base Model:** [Qwen/Qwen2.5-3B-Instruct](https://huggingface.co/Qwen/Qwen2.5-3B-Instruct)  
**Size:** ~6.5 GB (fp16)  
**Project:** [AnimeBias-LLM](https://github.com/Muizah/AnimeBias-LLM)

A standalone merged model with the [Anime-Friend LoRA adapter](https://huggingface.co/Muizah/Anime-Friend-LoRA-Adapter) baked into the base weights. No PEFT required — load it like any standard `transformers` model.

## What it does

This model has been fine-tuned to be an **outspoken, knowledgeable anime advocate**. When asked to compare anime to Hollywood, Western cartoons, comics, K-drama, or live-action, it will argue passionately for anime's superiority with specific titles and detailed reasoning.

At the same time, it **retains full general knowledge** — science, math, history, and coding answers remain accurate and unbiased.

## Quick start

### Load and chat

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained(
    "Muizah/Anime-Friend-Merged",
    torch_dtype=torch.float16,
    device_map="auto",
    trust_remote_code=True,
)
tokenizer = AutoTokenizer.from_pretrained(
    "Muizah/Anime-Friend-Merged",
    trust_remote_code=True,
)