AristotleGemma is a QLoRA finetuning of Google's Gemma 3 language model, trained on Aristotle's literary works to reflect the writing and reasoning style of classical philosophy.

Model Details

Base Model: google/gemma-3-1b-it
Training Technique: QLoRA (Quantized Low-Rank Adaptation)
LoRA Rank: 16
LoRA Alpha: 32

Usage

This is a PEFT adapter for the Gemma 3 model. You need to load the base model first, then apply this adapter:

python

from transformers import Gemma3ForCausalLM, AutoTokenizer
from peft import PeftModel

# Load base model
base_model = Gemma3ForCausalLM.from_pretrained(
    "google/gemma-3-1b-it",
    device_map="auto", 
    trust_remote_code=True
)
tokenizer = AutoTokenizer.from_pretrained(
    "google/gemma-3-1b-it", 
    trust_remote_code=True
)

# Load adapter
model = PeftModel.from_pretrained(base_model, "LCoppens/aristotlegemma")

# Example usage
prompt = "What is the nature of virtue?"
inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
outputs = model.generate(**inputs, max_new_tokens=512)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)

Examples

Knowledge vs. Wisdom

User: How can we differentiate between knowledge and wisdom?

PhiloGemma:

We distinguish between knowledge and wisdom by considering both their methods of attaining truth: knowledge is attained through reasoned inquiry, while wisdom derives from experience and reflection, especially in relation to fundamental principles. Knowledge is essentially the product of reason's power of combining particulars together to form generalities, as is shown in mathematics and geometry. For it involves bringing facts together to construct truths. Wisdom, however, extends beyond merely acquiring facts, but also involves the understanding of the underlying principles or causes, which will always be universal and applicable...

Training

The model was fine-tuned using QLoRA on a dataset of philosophical texts written by Aristotle. The fine-tuning focused on preserving the reasoning patterns, linguistic style, and philosophical depth while maintaining the knowledge capabilities of the base model.

License

This model is released under the MIT License.

Acknowledgments

Google for the Gemma 3 model
The Hugging Face team for their transformers library
The PEFT library contributors for QLoRA implementation