Model comparison

SuperHY3-abliterated-NVFP4 vs Hy3

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

From the file· summed bytes, KV per layer

Side by side

SuperHY3-abliterated-NVFP4Hy3
Parameters172B299B
Architecturehy-v3hy_v3
Layers8080
Native context262,144262,144
Mixture of expertsyes, 192 expertsyes, 192 experts
Quantizations published137
Smallest quantization95.10 GiB59.65 GiB
Q4_K_M168.57 GiB
Licenceapache-2.0apache-2.0

KV cache by context

the term that decides long-context viability
ContextSuperHY3-abliterated-NVFP4Hy3Ratio
4,0961.25 GiB1.25 GiB1.00×
8,1922.50 GiB2.50 GiB1.00×
16,3845.00 GiB5.00 GiB1.00×
32,76810.00 GiB10.00 GiB1.00×
65,53620.00 GiB20.00 GiB1.00×
131,07240.00 GiB40.00 GiB1.00×