Model comparison

rocket-3B vs Qwen3-VL-8B-Instruct-abliterated-v1

At Q4_K_M, rocket-3B is the smaller download — 1,708,595,136 bytes against 5,027,785,792.

From the file· summed bytes, KV per layer

Side by side

rocket-3BQwen3-VL-8B-Instruct-abliterated-v1
Parameters2.8B8.8B
Architecturestablelmqwen3vl
Layers32
Native context4,096
Mixture of expertsnono
Quantizations published1248
Smallest quantization1.12 GiB1.97 GiB
Q4_K_M1.59 GiB4.68 GiB
Licencecc-by-sa-4.0apache-2.0

KV cache by context

the term that decides long-context viability
Contextrocket-3BQwen3-VL-8B-Instruct-abliterated-v1Ratio
4,0961.25 GiB
8,1922.50 GiB
16,3845.00 GiB
32,76810.00 GiB
65,53620.00 GiB
131,07240.00 GiB