PaddleOCR-VL: Boosting Multilingual Document Parsing via a 0.9B Ultra-Compact Vision-Language Model — mentioned by meander_water

Sun, 19 Oct 2025 12:16:39 GMT

A couple of examples I have seen recently which makes me agree with OP: - PaddleOCR, a 0.9B model that reaches SOTA accuracy across text, tables, formulas, charts & handwriting. [0] - A 3B and 8B model which performs HTML to json extraction at GPT-5 level accuracy at 40-80x less cost, and faster inference. [1] I think it makes sense to fine tune when you're optimizing for a specific task. [0] https://huggingface.co/papers/2510.14528 [1] https://www.reddit.com/r/LocalLLaMA/comments/1o8m0ti/we_buil...

Paper and discussion history. A mention is not an endorsement.

PaddleOCR-VL: Boosting Multilingual Document Parsing via a 0.9B Ultra-Compact Vision-Language Model — HN submission

Sat, 18 Oct 2025 05:47:17 GMT

Multilingual Document Parsing via a 0.9B Vision-Language Model

2 points; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.

PaddleOCR-VL: Boosting Multilingual Document Parsing via a 0.9B Ultra-Compact Vision-Language Model · HackerPapers discussions

PaddleOCR-VL: Boosting Multilingual Document Parsing via a 0.9B Ultra-Compact Vision-Language Model — mentioned by meander_water

PaddleOCR-VL: Boosting Multilingual Document Parsing via a 0.9B Ultra-Compact Vision-Language Model — HN submission