CasperYL/pascal-unsloth-mistral-lora-en

Pascal Mistral LoRA EN

A small LoRA adapter fine-tuned to imitate the English "deep thought" style of Pascal, the relaxed sea otter philosopher from Animal Crossing.

This is a fan/research project for local experimentation with character-style fine-tuning, maaan. It is not affiliated with Nintendo, Animal Crossing, or any official rights holder.

Base Model

This adapter was trained on:

unsloth/mistral-7b-instruct-v0.3-bnb-4bit

This is a LoRA adapter only. It is not a standalone model.

Intended Use

Best for short, single-turn Pascal-style responses:

• casual island chatter
• surreal little observations
• "deep thought" style quote generation
• local LoRA inference experiments

Limitations

• Works best in single-turn mode.
• Multi-turn chat can drift or become repetitive.
• It may produce incorrect, strange, or non-canonical lines.
• It imitates style; it does not retrieve an authoritative quote database.
• Ordinary direct questions may produce more generic chatbot answers.

Example Prompt

Player: I found something shiny near the shore.
Pascal:

Loading Example

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

adapter_id = "CasperYL/pascal-unsloth-mistral-lora-en"

tokenizer = AutoTokenizer.from_pretrained(adapter_id)
base_model = AutoModelForCausalLM.from_pretrained(
    "unsloth/mistral-7b-instruct-v0.3-bnb-4bit",
    device_map="auto",
)
model = PeftModel.from_pretrained(base_model, adapter_id)

Training Data

The adapter was trained from a small, single-turn Alpaca-style dataset built from Pascal "deep thought" quotes and hand-written player prompts.

License / Rights Note

This is a fan-made adapter for research and personal experimentation. Animal Crossing and Pascal are properties of their respective rights holders. Please respect the base model license and any applicable rights around source material.