模型44

LlamaIndex 推出文档解析模型 OCRia-1.5

原标题:Today we’re introducing OCRia-1.5

Jerry Liu@jerryjliu0发布时间 (北京时间)

LlamaIndex 发布文档解析模型系列 OCRia-1.5,含 2B(Cost Effective)与 4B(Agentic)两个尺寸,价格档位不变。

中文译文尚未完成,先为你展示原文。

Today we’re introducing OCRia-1.5

It’s a new family of document parsing models that push the boundaries of accuracy and cost across multiple document parsing dimensions.

We had just released OCRia-1 a month ago, which set a new SOTA in unified document parsing, but we've only released even more improvements since then.
✅ Tables: Table scores rose from 86.5 -> 90.1 on agentic, 83.2 -> 87.5 on cost-effective
✅ Charts: 74.9->76.8 on agentic, 67.3 -> 73.0 on cost-effective
✅ Forms: +2.6 on agentic, +4.2 on cost-effective
✅ Visual Grounding: +4.1 on agentic, and +11 (!!) points on cost-effective

Together, these models set a new Pareto frontier for document parsing. Our cost-effective OCRia-1.5 is the cheapest and most accurate model for parsing all of tables, charts, forms, handwriting, with high-quality layout.

Blog: https://www.llamaindex.ai/blog/introducing-ocria-1-5?utm_source=socials&utm_medium=tw_channel&utm_campaign=launch&utm_content=blog

These scores for all our modes (cost-effective, agentic, agentic plus) reflected within ParseBench-2: https://www.parsebench.ai/

Today we’re introducing OCRia-1.5. It is more accurate than OCRia-1 on complex tables, charts and forms, and it places bounding boxes more precisely. OCRia-1.5 comes in two sizes now: → OCRia-1.5-2B runs Cost Effective, built for parsing at scale → OCRia-1.5-4B, our upgraded flagship, runs Agentic for the hardest pages Same tiers, same prices, better models underneath. Here’s what changed on ParseBench-2 👇️
查看原帖
多模态模型发布评测基准#模型发布#多模态#评测/基准

同一件事的其他报道2 篇