lucasoc/sci-image-models

docs: atualiza metricas finais oficiais do conjunto de teste (373 figuras) e tempo de treino

feat: upload qwen2.5-vl-3b native original resolution weights and pt-br model card

docs: adiciona tempo de treinamento e hardware a tabela comparativa

docs: atualiza explicacao detalhada da loss icdar vs padrao e remove emojis

docs: update pt-br readme with detailed loss explanation, lora mechanics and +pp metrics

feat: initial release of sci-image-markdown Qwen2.5-VL-3B LoRA adapter and extraction codebase

initial commit