AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

TikZilla: Scaling Text-to-TikZ with High-Quality Data and Reinforcement Learning

Christian Greisinger, Steffen Eger

ICLR 2026 · Large language models › Instruction and fine-tuning · approach: Calibration and knowledge in LLM tuning · 5 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add RL with domain reward after SFT (ablation)Qwen2.5-3B · DaTikZ-V4 testAVG: 0.324 -> 0.385Improved
add SFT before RL (ablation)Qwen2.5-3B · DaTikZ-V4 testAVG: 0.269 -> 0.385Improved
RL directly without SFT (ablation)Qwen3-8B · DaTikZ-V4 testAVG: 0.251 -> 0.357Improved
add RL (ablation)Qwen models · DaTikZ-V4 testcompilation rate: TikZilla-3B 89% -> 98%Improved
use VLM descriptions instead of raw captions at inference (ablation)GPT-4o · DaTikZ-V4 testAVG: 0.270 -> 0.315Improved
use VLM descriptions instead of captions for SFT (ablation)Qwen2.5-3B · DaTikZ-V4 testAVG: 0.279 -> 0.289Improved
mix captions with descriptions in SFT (ablation)Qwen2.5-3B · DaTikZ-V4 testAVG: 0.324 -> 0.313Hurt
add LLM debugging of uncompilable code to training data (ablation)Qwen2.5-3B SFT · DaTikZ-V4 testAVG: 0.288 -> 0.324Improved
use domain reward R_Sim instead of DreamSim (ablation)SFT+RL Qwen2.5-3B · DaTikZ-V4 testAVG: 0.474 -> 0.498Improved
retrain reward encoder on DaTikZ-V4 instead of V3 (ablation)reward model · DaTikZ-V4 testAVG: 0.487 -> 0.498Improved
increase SFT data from 0 to 25% then to 100% (ablation)Qwen2.5-3B · DaTikZ-V4 testAVGImproved
evaluate TikZilla out of distribution on SPIQATikZilla-3B-RL · SPIQADreamSim: GPT-5 0.432 -> 0.637Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy