Tencent's HyOCR-1.5 open-source end-to-end OCR large model

Tencent's HyOCR-1.5, an open-source end-to-end OCR expert model with fully open-source training and inference capabilities, is the industry's first. The model incorporates DFlash speculative decoding, achieving a 6.37x speedup inference speedup under Transformers and a 2.14x speedup under vLLM. It achieved the highest score of 94.74 on OmniDocBench v1.6, ranking first among top-end OCR models. HyOCR-1.5 supports eight task categories, including document parsing, ancient character recognition, chart parsing, and multi-page question answering, covering 331 languages.