Model comparison

PaddleOCR-VL-1.5 vs ced-base

These two publish different quantization sets; the table below has the exact sizes.

From the file· summed bytes, KV per layer

Side by side

PaddleOCR-VL-1.5ced-base
Parameters959M86M
Architecturepaddleocrced
Layers18
Native context131,072
Mixture of expertsnono
Quantizations published223
Smallest quantization0.25 GiB0.12 GiB
Q4_K_M0.28 GiB
Licenceapache-2.0

KV cache by context

the term that decides long-context viability
ContextPaddleOCR-VL-1.5ced-baseRatio
4,0960.07 GiB
8,1920.14 GiB
16,3840.28 GiB
32,7680.56 GiB
65,5361.13 GiB
131,0722.25 GiB