Ti
kZero+ (10b) is a multimodal language model that automatically synthesizes
scientific figures as editable, semantics-preserving
TikZ graphics programs conditioned on text
captions. It is based on
DeTikZifyv2
(8b) and
LLaMA3.2
(1b), and integrates
TikZero with additional
end-to-end fine-tuning. Check out the
DeTikZify project for more
information and tips on how to best run the model.
1from detikzify.model import load
2from detikzify.infer import DetikzifyPipeline
3
4caption = "A multi-layer perceptron with two hidden layers."
5pipeline = DetikzifyPipeline(*load(
6 model_name_or_path="nllg/tikzero-plus-10b",
7 device_map="auto",
8 torch_dtype="bfloat16",
9))
10
11# generate a single TikZ program
12fig = pipeline.sample(text=caption)
13
14# if it compiles, rasterize it and show it
15if fig.is_rasterizable:
16 fig.rasterize().show()
This model was trained using computational resources provided by the
bwForCluster Helix, as part of the bwHPC-S5 project. The authors acknowledge
support from the state of Baden-Württemberg through the bwHPC initiative and
the German Research Foundation (DFG) under grant INST 35/1597-1 FUGG.
