Much worse than old model on llama.cpp
#11
by
engrtipusultan
- opened
I have used Nanonets-OCR2-3B f16 it cannot properly parse your invoice samples that your website provided. v1 can OCR them fine. On long text images only last part is extracted in v2.