dots.ocr

• rednote-hilab/dots.ocr
  • dots.ocr: Multilingual Document Layout Parsing in a Single Vision-Language Model
  • by 小红书
  • Layout 分析效果非常好
• 参考
  • https://huggingface.co/rednote-hilab/dots.ocr
  • https://github.com/sljeff/dots-ocr-client
  • https://replicate.com/sljeff/dots.ocr
• DotsOCRForCausalLM
  • Qwen2ForCausalLM
  • https://huggingface.co/rednote-hilab/dots.ocr/blob/main/modeling_dots_ocr.py
• dots.ocr
  • silu
  • hidden size 1536
  • max_position_embeddings/context length 131072 128k
  • 12 heads
  • 28 layers
  • 2 kv heads