Model comparison

OpenELM-3B-Instruct vs Llama-3.1-8B-Instruct

At Q4_K_M, OpenELM-3B-Instruct is the smaller download — 1,887,788,096 bytes against 4,920,739,168.

From the file· summed bytes, KV per layer

Side by side

OpenELM-3B-InstructLlama-3.1-8B-Instruct
Parameters3.0B8.0B
Architectureopenelmllama
Layers32
Native context131,072
Mixture of expertsnono
Quantizations published1545
Smallest quantization1.07 GiB2.02 GiB
Q4_K_M1.76 GiB4.58 GiB
Licenceapple-amlrllama3.1

KV cache by context

the term that decides long-context viability
ContextOpenELM-3B-InstructLlama-3.1-8B-InstructRatio
4,0960.50 GiB
8,1921.00 GiB
16,3842.00 GiB
32,7684.00 GiB
65,5368.00 GiB
131,07216.00 GiB