Modelpublic

lee851104/gemma3-4b-astronomy-lora

sourceHugging Faceotherupdated 1mo agoView on Hugging Face

0likes5downloads

Model Card

lee851104/gemma3-4b-astronomy-lora

LoRA adapter for unsloth/gemma-3-4b-it-unsloth-bnb-4bit, fine-tuned for astronomy image captioning.

—rank r=16, alpha=16
—trained on 200 held-out-clean records (split_hash e2d16e46547c6f61)

Usage

python

from peft import PeftModel
from unsloth import FastVisionModel

model, tokenizer = FastVisionModel.from_pretrained(
    "unsloth/gemma-3-4b-it-unsloth-bnb-4bit", load_in_4bit=True,
)
model = PeftModel.from_pretrained(model, "lee851104/gemma3-4b-astronomy-lora")
FastVisionModel.for_inference(model)

Limitations

Not for scientific interpretation. Trained on ~250 captions for well under one epoch, so it learned output style rather than astronomy knowledge. Full limitations, known biases and out-of-scope uses: see MODEL_CARD.md in the source repository.

Source: https://github.com/lee851104/lora_mars