Model comparison

NEXUS-Legal vs Llama-3.2-3B-Instruct

At Q4_K_M, NEXUS-Legal is the smaller download — 986,045,472 bytes against 2,019,377,440. At long context the gap widens: NEXUS-Legal's KV cache at 32K is 4.0× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

NEXUS-LegalLlama-3.2-3B-Instruct
Parameters1.5B3.2B
Architectureqwen2llama
Layers2828
Native context32,768131,072
Mixture of expertsnono
Quantizations published2339
Smallest quantization0.37 GiB0.85 GiB
Q4_K_M0.92 GiB1.88 GiB
Licenceapache-2.0—

KV cache by context

the term that decides long-context viability
ContextNEXUS-LegalLlama-3.2-3B-InstructRatio
4,0960.11 GiB0.44 GiB4.00×
8,1920.22 GiB0.88 GiB4.00×
16,3840.44 GiB1.75 GiB4.00×
32,7680.88 GiB3.50 GiB4.00×
65,5361.75 GiB7.00 GiB4.00×
131,0723.50 GiB14.00 GiB4.00×