Update: I fine-tuned Qwen3.5-0.8B for OCR and it outperforms my previous 2B release [GGUF]
Hey everyone, A while ago I shared my fine-tuned Qwen3.5-2B OCR model. Since then I kept working on the pipeline and just released a new version based on Qwen3.5-0.8B. This one uses improved training samples and better output formatting, and it’s outpe…