svg-model — Text-to-SVG (Gemma-3 LoRA)

A LoRA adapter on top of gemma-3-27b-it that generates SVG vector graphics from natural-language descriptions. Give it a short prompt and it returns a complete, renderable <svg>…</svg> document.

• Base model: unsloth/gemma-3-27b-it-unsloth-bnb-4bit
• Method: LoRA (PEFT) fine-tune
• Training data: omeryentur/svg (68,554 description → SVG pairs)
• Live demo: omeryentur/svg Space

Usage

from peft import AutoPeftModelForCausalLM
from transformers import AutoTokenizer

model = AutoPeftModelForCausalLM.from_pretrained("omeryentur/svg-model", device_map="auto")
tok = AutoTokenizer.from_pretrained("omeryentur/svg-model")

prompt = "Abstract geometric pattern in teal and orange"
inputs = tok(prompt, return_tensors="pt").to(model.device)
out = model.generate(**inputs, max_new_tokens=1024)
print(tok.decode(out[0], skip_special_tokens=True))

The output is SVG markup you can save to a .svg file or render in a browser.

Intended use

• Text-conditioned vector-graphic / icon / illustration generation
• Design-assistant tooling
• Research on structured (renderable) code generation