Model comparison

LFM2-2.6B-Transcript vs Llama-3.2-1B-Instruct

At Q4_K_M, Llama-3.2-1B-Instruct is the smaller download — 807,690,688 bytes against 1,563,669,248. At long context the gap widens: LFM2-2.6B-Transcript's KV cache at 32K is 2.0× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

LFM2-2.6B-TranscriptLlama-3.2-1B-Instruct
Parameters2.6B1.2B
Architecturelfm2llama
Layers3016
Native context128,000131,072
Mixture of expertsnono
Quantizations published1139
Smallest quantization0.92 GiB0.39 GiB
Q4_K_M1.46 GiB0.75 GiB
Licence——

KV cache by context

the term that decides long-context viability
ContextLFM2-2.6B-TranscriptLlama-3.2-1B-InstructRatio
4,0960.06 GiB0.13 GiB2.00×
8,1920.13 GiB0.25 GiB2.00×
16,3840.25 GiB0.50 GiB2.00×
32,7680.50 GiB1.00 GiB2.00×
65,5361.00 GiB2.00 GiB2.00×
131,0722.00 GiB4.00 GiB2.00×
LFM2-2.6B-Transcript vs Llama-3.2-1B-Instruct — size, memory and hardware fit — ossmodeldb