Hey folks, this is for those who appreciate experimentation on open-sourced AI models. We fine-tuned open-sourced SMLs (3B and 7B parameters) with SFT + DPO against commercial models like GPT-5.4, Gemini 3.1 Pro, Claude Opus 4.6, Google Document API, and open-source alternatives like OlmOCR, Deepsee
