brighamb/PaddleOCR-VL-1.6-Q8_0-GGUF

PaddleOCR-VL-1.6 decoder, Q8_0

Q8_0 quantization of the PaddleOCR-VL-1.6 language decoder, produced with llama.cpp's llama-quantize directly from the official BF16 GGUF at PaddlePaddle/PaddleOCR-VL-1.6-GGUF. No other modifications. License follows the base model (Apache-2.0).

Use together with the official BF16 mmproj (PaddleOCR-VL-1.6-GGUF-mmproj.gguf) from the base repo.

In greedy-decoding OCR tests on typeset book pages, Q80 output was byte-identical to the BF16 decoder (including on long structured table regions, where Q4K_M showed occasional single-word insertions).

Hosted for on-device use by the Bookery e-reader.