Add rendered images archive (15k PNGs, extract with tar xzf)

Add RSFT filtered JSONL

Add HF Arrow dataset (train + validation splits)

Add GRPO Qwen2.5-3B full model (DINOv2 reward, 1701 steps)

Add DeTikZify-v2.5-8B SFT LoRA adapter

Add SFT Qwen2.5-3B LoRA adapter

Clear repo for reorganized model uploads

Upload best text-only SFT LoRA checkpoint (Qwen2.5-3B-Instruct)

initial commit