Model comparison

translategemma-4b-it vs HyperCLOVAX-SEED-Text-Instruct-1.5B

At Q4_K_M, HyperCLOVAX-SEED-Text-Instruct-1.5B is the smaller download — 1,133,974,368 bytes against 2,489,909,312.

From the file· summed bytes, KV per layer

Side by side

translategemma-4b-itHyperCLOVAX-SEED-Text-Instruct-1.5B
Parameters5.0B1.6B
Architecturegemma3llama
Layers
Native context
Mixture of expertsnono
Quantizations published191
Smallest quantization1.61 GiB1.06 GiB
Q4_K_M2.32 GiB1.06 GiB
Licencegemmaother

KV cache by context

the term that decides long-context viability
Contexttranslategemma-4b-itHyperCLOVAX-SEED-Text-Instruct-1.5BRatio
4,096
8,192
16,384
32,768
65,536
131,072